ࡱ> nmٮ5FE3]u7zn PNG  IHDRS{ pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14"IDATxq0 E-"TO~hP"Ex\8 y=(@W$3Fg9Bfñ[o4.#NqRĕr2MJsqŁO(H>Ek@MyN%Iqܷ32?#Nθ$b|c|kG>ǜKA 5PG2r{ KNIA˔''kC|0A=@R)a{)5ZFQ}J^ K'wM?y|#^KХN'sF"!Coa3MD=;Q>VI4}kfq&&ƣoGAT&zVK}7wWmt1Qw]]+ l ʹxh yT'p6?}LAuUO MytwU%CIoʝ,I+)泗*В#^^R$K!E%565},)lM%zIOJ/Wz ,ܹᣒ4^+)y-"4Km; kK#% B\kLJ  Γ\3F[RR!%%e?لgutIENDB`nd xa"NTyݨIPNG  IHDR*Aϸ pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.13lu IDATx]rPͭ,f^]>JQD/Ѝ>LQD^q- $k0nL?CwB/ENI(D>*IP|y$sg*" *" *" *" *" *" *" *" *" *" *" *" *" *" *" *" *" *" *" *wϏՒ%r, !B6,-\\mൗ-}Qᓞ}A./ۙ}2c# Fr #Mg#@@&Ge7aK$Q Q:1v2Ʈ dcH}b!~&^ \-G=9LM{"E|. ~]adKYM2!Hp^I;qN8joԫƲAUyՌ Α86^|Ot@0p@:gQv# %H]3 qY5\U-<KN Ͼfu5v%S)gdgki&R8unCK~~/7}^/}ԟl@^V؆F~t`CAa@~9D3? _;0`&sAE;{~}1O,7g$2: 6f>/Yr@*@I'cfF̓ڛyZ5HO(E;Crn{@,埣~ڌ}F;:CAa\g3Opƌ\V|ofo{i\?kϓ ]6 8Xyz&@XN,#A1\Fh'ˇfVwD?R `.ogsh^C(jmd}c\Gjp +r2IBS2^ x|7IĹ|]i !ȘS*B^׶ƿEc$0mE1 ‘UDM%z({eu%fα'Am~L :mIlpڼ|[wVBml~)0Aqٴ!q9mHj7M99`řl&^@Op뷙t9 a/+ij xgZb.w;MI_ZUU?`~/mSH/x}_Y) ٹؒ\i>i-s@b{w+lg z}21mF@ϥU%Z I}b3 rE063w5tK1xGt #oXM6.pTVf_'@wHM1~W;m9@`nQ3H@kw>Hl 8% C^ɊE7Kڳ{y.)Ŭ$u, a' ;/g,J-H a?Uv/EVmG8rgSS< (WIЬoO-ύ4B%SG=  8e\)fAMG;&eY9*0Qv7i>%6+ཀn$6o^;8͛pԇs?`4h!у=̩EvjzC4@8 Z@ bFc ruD@umf_ H4#Բ~ҕ@KC➀GyWƂW|^>)xGӠ+ [/ON $j/QmLK"ljOTW߯i]ԉXV*aw.:|PcMS8K1_a/Gڟb9  K*똬)PgQ&nI .êRcn9ָ%ۖn7͏W uRv:ǯUa,:xZ"m;;~"=}{DȽ}h@\t/K o(? [[y:_ȏxeD"H4~Hđ0VCcv Jh}ིˆU6ZΎWh6I+|v}9L9/\v <ӌx~$q{;na0ùf.ٮ ?דw5{~ӏpN}6җS>^X<_lKԥgGfሲpv݋?4i Efo2@l˳"@?-R;+y}E(pwEG.bc=){1c]:}'}.r;9EއVQ_VBb-)m1ٳ뵞ed;]TX3"`V6Ehy/1ן>n9+~h7r6fNppbo^?)+{Aێ` wq1ȱ5FگuY5>O56!,#-gGE.e 7s6~R 絯ĀsVjZR]Sv4n|"ìS8S@d崙F]vm {_>r8NjRoѮGLl\n>^ czRJ+-gEh29[F D/+ё)܅tQJo=io\zԻwQІz*(#j 0ZjeW%GYLZvf/>KSьganE3U/sxǿݣ<)UOiًs5-v:fOh?;29M6ГNudQp|lu8㑬^t x_'e]Lx@3>#I{T72:2.Zbgv m("lÎ(x 9^̠dSR"WK^Mr'-A &Wb"ԈR"atwޗ}:.X㰻\ 4IO(Z]c1~&0n0>mLf@ MRӑY{Ny@@OqS%jnk ]p%{[F֜,E*%o=v/xuj1qٓ3_;=yz1 T5Dvjv9Zӆ׾`jb\Y`٥p+I-λDyF8mD׏ Rq3@.Z<ì1 Ѿ.d$hЕ.tPvדN s~11A<10͍Aq)-q"w 뤋ڬ$eEBm.r%kspe̎qɜ xiE2s2z>[td {BELng^t8.ͬ :" 41;"Ke^^\Q+LIٓzҩYxqWK}z)U_oŤ Ȝ,Q){LORrpʘ*Ɲ-I1_+Iٶד&qp=\.c8j:" ث\)}t6nifZIZO򋋰쟅(kUbOzW[ ;;Њg PMUH:ƀU综cPSx>Ee-ZOJ Ї}ݮ8*)C_R,}vfbŤq 5~%s|S3XbFrY b`S^Y,29'i IjlQJlNj1>UuoV"'!s5zIt1gG_8/&%1@q#s~[r=#=c O4vxݴ6A8hʘ.ʸC?JRZv||zR-xCoNSo_>'ex_{^ W/-+Tu\~]QU]Q!|.6WȜ 5A#y 10}9^(TrjNFTFo?AXO`QXm9|]>2K;m9XO`i ﲗFpoz7z:zY?φnSހFo?}ٟ@OJ@ I8IXO ZE vIENDB`n-qqԤMPNG  IHDROղ pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14"IDATxu#9 @9qTr>B>՚Z(.~vWX@.E >_?M@PQXr['*HI7m6_Մ,i  0}&<']1ȍPHۮ,@Kq;"-I.BKyr-ΛF(TmSX~~x6O XA䇉ng CG&,mID =}0Gn5 0xD 3,bo)Y8;VF(ɧmC$l"OObz) K ,Kwf#ac.B,msN6#`Yiu7,vUnE(4zuv%E(ą9I^f)e\m &_ @(6!7h%LOڶ-Č7g'əfV[tc![AMW}d8_Aws_Y8hE ?ӶnĉQpaë́, etw_qSJ[(exq==ӻ:Xʋg!C`woZ5.p,ޝ[`5+h<;[vK TK! wLڑn9֘*p 찧>J崬o=ruV5}Q([[ä@. 1 ISz-{\wkxo/Ι~O.3aMJw& aÉEBAi%X֐"={ FeJORGCw4KBwfp< zX[z@!mtq·Pp'܉Uu޷TGj WK.\q=y1}K/`SZCyV!b.Tjmw!j감넂hd:8Fy%VȍXUUZ'4q+ll{F+sC⚚}z|h-oy-F["Ss5I@(x5_0=7lQU;jo:0ƞ7j ̸mYPeQe/ '^[ϕݣB/E gH&T '`Ԗ5yc"B[jKux4m5\VC%B6ug?[}(ۥwi|E;A(0AbMW7}sg5k*ޤmQ#'@wcD*Ԍ@SJz&V:ygE()9v O;F7ywgLz4eMC^8DI+g ϙ{]b]׎t[֗I)b|'UsX# ^lXyh6L4DTK o^(qPpR T]ʍm_mkWsO­M9oVE(kʮ@Cܷ o%eDĵx3}sDeA$$JUP<|z>A@1}gH۰jϓ"Бr*mX3 fUre m,UXFھ0r"e*Y֝ YP v moU`iS_IHbm WAE# \iw;Cd;eˏT .jS1w<^~?!j𸕣֏'r6CʳhRUI \ #A ǂ2Me}A#XgNszS~=_ 7JFbzG`/emڮBnxj0-YݑPIENDB`n ,:u^J]PNG  IHDRdH pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.13lu QIDATxu8 aϞ-bRAJ?"YődH/;) a~}}}Ux7h`4 VH׻~O " v1=f5vdaݱb"" :1afuqdqGɻ=FDKLY YvX0#[ Y ֿ-0Y5F#(j֬,gX=@*ft;YlW5͎v7^e|͝77q=/vXk_n^(+ >bz~fl1bMyg;:r0huG5.u4ekj|N#j\Hyd\5OrI%]!C#[ 1l7$Y1 F#r!$ ?/]Geg?ģo-T@b9s췯Xu쎬kp1A˽% Xê-E`]xXi#*PS{$[<5Α=V*PqֳjilotUȖj3V2k :cSbM7ͪnޝ#ta?Y-r?0@nj .V($^?>/7wWp<ښ$ E7(o$39KaPhX5uCЍ7*D u 0IY߄{5eink20QF=>U[McdT݈5 zqQf2fÖ5>ymK̠Z-5c(a`垹;]s:A8Q&!U{B(XKNm[ӶXc~TL&ʕ@?$s{8ﰺ0A7ZuqDDn(X2 :jԌQb/]4lf<<8oXsKm*c%j\ʨQP 0{cΌJœKjx }hf<ϰ*(φ((͡Gf?OȲpZkUʫZ-5cLXISmΰcC.RmoHآOpjuy`նS3! w)Jv9jG`M*$H16v .;鑘 4Fg͍KswwX9銡]\YWVW%ռX_~L^X9kGسi^t6V*'W}@ T[w}nͻyլVEN >U۞];-\m 6(KSMp>Ĺ$KYT"%9>a٢w7s B=n+a3'GqAo.#"ժElY~ӹZDUX0vw9@. [G^{\NVoϷЯV)byU2`\ߵge^}Dzyj5aܚGj sGy9/] kVg͎TWȺ|t4c̋gO}Vm}?ΰ|wl]1SCeվ UOUjه 3OνHs*k4ncPU%1)It8jes%`WveFzg[,}cok*o/B_ y4C~ݣ]ӥo>*AGCd;(Ȫ-k§a"tԼMxn5~rgl <0.kg Ma~ :LI f~Ƭ/',+F(,gGkƈ/_~ H0/U%Vדy[fl.DZS#*3rE8~v2w}yvZ%u F?ۙȶR&{e Fc܇#PBr2 ]ԈwV8('bd(mH Vz`/wD! rT8 r:ي^| |vU(DA#"m}9E ]#D )u`btT HJ?!.jv#kʏqC\ZY9JGV!/kc n%3U2I4쏿0J'ܔr;zD! reV+#[w.ěWhiHQ'h#p2V/Xo*ƅK!$=DHTZ*s,FaxGaQ"h#RE2wR¸yP; 2EGJT;jd)X0}(M ]Bs)V f`sۏ+ 6n~Ʀ(D!]"@hٝ9sI;.7)U/+gN#6{uM(wlt#Chc[6J{SEp>r~u2wl(XZ@DdQa,"'r#3Fx7 De.+]? "r(}Rb OLBg._Žv 2!*d.;x􃚪TH"ʝXk?DЉ"*wSoQ.軴9 e.;8c7CD<+} K i| ;`]*IENDB`n?H8$^3CPNG  IHDRdW}nw pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14"IDATxm0 ECdL 2U釁ĕeHq||EӒ ?Kp~|<_&AVƏ5 hmdYO dUЅek{{YNeXV/j+Y[=,[̻eRV_+@% `(Yx<6<V =SwYqc>bD,{oWK0ecvWvG:1 <Q6o :wLFu) dy,-BΉ(jJJ5iƕɺ߮Q;.MiR WZ,LVvzwC8md1/i+dEj`uIy:5KEd@WG'mOr$InQ2AeISDYB9xNr%@*u=Yi^J}2;f/gX~Cl${2iYїN@Neq:i5YJO:;Ӕ셄,싺cI, a+Vh-L+ AZ6cY zA``ZP&kXUEeM޵YJqb!Kz3UdLm5siHjxd /Ci'Va Eɚ*VDl™,$ԛH$/`2>|zIENDB`n ono * PNG  IHDR*j4 ! pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14" fIDATx]rPы`?^WUy? T?UD?Lt{$T)8NhO `F{o#<ߪ %u9>8~ l~7k?<3?s2}yuHMTy6.n=Ur3JCk<y=eBZέIe&~ ';  - |_Gs%Q~HrAPk)'-H zh03C#E ?>iʼn<\ʘO9 * V1ZAP1-%`0⇒jOaTlH} LNȣxży 2_M,U;H%Ga=5]!i¹ɳ - P7| "}f '~إHNhJ`BдA0*Cۗ!x+;3[?@$ C`C, gEG{IyDj%Sġᵗ5X%,q|Ntt?t ~@Ё@:?t ~@Ё@:?t ~@Ё@{o<ބ~{oi7r9=%6`{"3mD;B/.GQƽQ)ө ;oq^k=!!u9u !8B(/_ zC,kh pu>sb$~Hiz@#L2䮭IƣkcTm"t慟ɷU{8i_qzd& LwjU٠. h"P@G3NCN8T3o7ϓSgq|6uEX?:,^?-tNգzI̞ޖݹƣ,΁@HQ="lK6Ƽx[y\4i79!7û$A=4M=l3,rƓT KZ_0f,$~侔!`T>vJ"~Ayc6dLHЙ9h}{:?t ~@Ё@:?t ~@Ё@>@l:{:?tp]pgoϵðCkk`ǐ~hjnNx-(?[? Yc$&ߠ=rfKM}"l,@#Vn4@AЁ>LBdZ=?ft`7?ts?u&15}]Kp`ECRx\ JQU`/)g;!CyZ<{k5ٳ^I, qX!:'6= LNPXOP7` CfwQPP75Mp?ԦMç$CSJP[Iv|"%2E.l)g;!Cg|vHE:ZJڸ!,, ECo]O0&9 F@&'3oS]jNHA@S &L [GzOs=q:zoc7r96cZx~'?dPѓQ@/ܤΜݔWiTX@QܩeIENDB`nz 4!+tCSDPNG  IHDR3>! pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14" IDATx۵&a+kH?SU*HU']8^Q$q7ޒ9F -$x/pw@?+@q^ML{ϛ}N5)BXkx\1Y1yDސ$ B0Lr]nrmd&S,,8[h;%.c!&d( l`d& r7ƽoR5ʵq7)|A( @;23?2L#3ZV׾EW.9^8/;e|ffs-FGU%_HD3b2u8B떯O뇤刡m7Xԧ3Moo#O$:R/A հ~tT@&RQF,fo^z ҅&zu>neMk#]8NTjpvBrDx ڶu'[G(b- L4sz-4fKex~zRi7J&w,Q12:CC-gT*-)Gi:-(J+ǭɔԣ.su[,S<R²+8yU߭DOF%r~{0^ZoϜE Tz .w.6teh]Yo 2bRSm58[e8E+Sm92rc֓-aGGe f^SԴX 3g _wuԤkxG!(jMoQ@,'hmpęGYݪEuJHK*jd `Z~t/Μ;!3l%5x@fj#i ;-6~&&a+VCS6Lm9vdfґcXZVLqJcE)=0nΜ^kf}Wi /F#,8Dni K&ܬb-0:DCa,|G_?XD'{rwne"zS/ ?g&W{JK_/2=;`+)#~ }%Z! n<> =n`7_ %-!@0&\}, |4:jV /%>od9hřH3:xG"@ZU2K^A-Fֽs{rREL 5)UoK%.x2 `#od9;Sgf#p(b'_I~ D;d=#@9ifr '@J7%H@0PL`m|>-!@XCvtu4=?I O;5ܖbH9hZ\y$'{bRX̤)1O7R~xPvrc"% `挬(m\u'P q2:; fѳgҚTpS#|emsV岟caǸPf&/R]Ϲ}|3?^%Gq5vFkǡ`p֑4NK\s]l$%@.fJ/%>#݅~I l@UmvI9}')G?tiISf΢S?l|#Ͼ/8x; Ooy6J8U~w&5#bO*rZ'@gp9mմ(d&wz9ib>W]TPjfUm*6髷UmR/#8W ".yX}SDw(Ԝ()yLEIsP s.sF|}VݼI8Rt>S(B*r̙g|}99 q4#m~B##Xڼہ&u8:҃ z3gg3{jvE|Ӓ7*$^^ciuݤ,X>c:̈́#j+@4oGqku q_BN:$#̜0"#"wBtb0s''%a33a}Knjd#]qdwKg>+89{Z5@bSBSiW F҆D*s'9 XXpG8dEPM3g%NkNLIENDB`n( VJ^]/o]uPNG  IHDRO"np pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14" IDATxu8a+" !i']^c˲dH3}XoH<N>Ө%/O`Kx Ł>c_eZE!\y".-OFG%Tl c! !3iTMSKӭ-fK(f9!0$YB@ _~a%WQj n4?OTBY7Li&$ZZQ3qajD}>A6}c&w՟@ޢ2k1X31x'>3I\8Cit)yD-TbjZЋԊF%`!,>e2n<%ceQK@]*i Hu`"#h`c6j ctjnZ%bwc,W5N7j hMK0K5qX7eF-6u` z j uܴD*AJU-[;e8/;Щ%!&j 8s6Xbl,1Fkzt)࠭EIsI.Nf-U Cj`Xa I%n5~B_߁>iڴ+q(юPrÎ㠫q .i 2N|%ۧG5sYT*՟F8xJܐe#nR#60oh豛&ْݼC{%/?MK#6-펜8|-FiTÛ ׸`MK˫y9яC#3ikLҚJ0Z^4Җ;_?v^) l{'=-t?C<ٚJ0Z^?$\KJmΐk:h͟;hcIߴT:+zE[ArW@,bjUÌ("^]9mlNG&eպ㓾mjΤ0B6-ALi):9tcN&6[z5GccK.ە2ȋҍ-7tDx잒%b5r$K2)˩17lZIGTz" mHU8F*1=z]f 7>IGKcz4rNc\&ˏo_UJ,=A!F:OhI#˃D/Tt4Bʾ艎u9.^Hc) }$zqjY:Ү˓9G~3)F ژ /үK00D0 WPb/qSj 73X.gH,@5` dXpȇhZKv^=qwXeZbaZa0XeQ.k)*zmqMElZG鋕}8wPK4FUMK0etPK;?qqk(㎋eFD\eX`i )`58hY-%,R)e8K. Z f>Tmk<&N~TS `n4ٯtJ,tBLD*(4ٯjrTS\5\p57F 0.S2,u'5$_U`kLtL5"ZhUIENDB`n8sb3ePNG  IHDR2%- pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.13lu IDATx E!Ot zDQb R)O3|m;z?6ASk''DJqsJeTH0իJ `XԬ-ңޜŻuf2D,¸ar59u$V >\m=e[l[lZ)-汮-,Zwq czM}g[ cr=%pV'*Gg>R ) eO#z'?Rކ1-"IENDB`n. Wh~tPNG  IHDR"] pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.13luIDATxQ:@QU QeUYt[-Y+/2`nЍ\};ǟJ@ IzgTؠOaR%/E>>V%/E xp%}J@k+RR%7Բ|[*Q%A)PIowEFroIިpJRR%7T?(%UNW;U5(%UzJ.ߒ*U]%UzJ@-RR%D*~ BG>۱7r- l)\Ps#|n?8n,Mٻ-əY RfV<±ٔ/ȟeUDY˫u=~Y6%6GT k3%H`H}"II`,+QT f0P*A$XIJ#S%~|F2rY vu"U'l:NU:t-D5uO'%*:q΢JP8RJ@JaR%`__ N** ,@ A`*֏so~Ae\Cc[hj0(gޖ{~^/$)M[?z,^gYh4S8b_v]:I1}nkkV 3HKO#OmL4+IC-bkR;pbfSﺖrJt'LJ"%.1UgdX $r¯6 #\w{G*< 'f]y=bV IgK*`<& $`EҴG9ТU ]*K+TT K*YcPqpεs?yT[M<8y OlBa%U*;6x߫wݓM slK+ٽ|obCJIc?2afWklGvr_lS? {k=Sr_[<<6/7I_ Ȫґ#g}7vH-]>}ʋ8nř?eRW.|`Y{y|࠵*UMҧAr-U:wg%1ݎmfEqLWpRЮ(vA U;pǹW)sLL*=-7_F];>sS]R6{S?%JyXyۉ!=$@*%yN*-=.:JS#rܻЂWiQq)1U v#UJ$I3ER0I!R.LDTU) $\UJ!$I6]a$‹y΀LÝc; fFߥGhY Գ!UJ@_T *}Q%/wG>}ޛx"+u'CkI ͲIäJs\\,jJ\ G86![̯kp0G_T*A4+d~Wβ*T b*KIJ:IRR% J0!JPTR*׉tXb D̫l\UJ@_T pA˷J0n_S% WWT (WcR%ow*ݩ̨`uPKYT ض/'J\JXJ0Ai|\֏so~:UJYA)̦QST 8+z> zyYv O^-g:eyhVq{+W^Yj ЗiCwIENDB`np gL,)i)JPNG  IHDR4 pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.13lu IDATxq:P{p>#pTOg-ZO DXֈC%wkGs^O- ~WNNS.;>|XĠ`%+F+:=M+J!HP܈xfٕkVY =`D:W2\@nұY<^Q7Oޟhf,7R JP'Zײ JRS p`"arp>%5Mg'޺s}꡽,5/^^r%|^"u`a6i:@{g|_ ~4?W*;N-n P ٰ˷4iFo{XS vWV;b'`g̩Ce>K~_A5.WkSspoxe؏٪Wl1rE$ѲP uL,ߡ{jl72{Qp<6]ޡ9˨eͯ",{$G՟<{kƏ~+|Y:GrHOJ?|kxۻO~UjGj\ݙ@=y+-r*\ \0Xrp"W$W$WL:2&W`F΂W.sR=wf,-rhA)g-j[Zg̃lS;ȓ?*5Ԟ+@ ˖:ctmovPz0(S(k*lIuY/xcvVΪk/ڗuW7ݮ\j)?rj<;e,z̉ /\ B7T亏+@^* v݆2Z\SJ"% WYn}8cw P׌?_'q+ $1i4{t+@E]^5-O7l K4&Zq-{7ԛW 0#Ytp۽m yt 4ek0]`b(e/Di<0ICN]s 'DٸR@ʑ.{@15' \FNE.0 r~+_,~wgy&WA}ȳ ;~`t:nS~AўkUi6*WvЍe:aM RRV,Oyt̘1$+ۭ[2vvww o`6ʘj`壩PGtuu:ej^/w%m ʋw| V6ctd@K޽{zt1P0FGGۃɚJJr]Ũs6 tі@ON|3788ST(,V,A HLsu. 08ww7 Z[Z]]֢ ǀ w=0jnnc,0Q X  b`) [P^yRՑq:jk a֡N*{rO?'M$CgϞoni% 66c<<\/Zd~It}kjhn.γ&%_<=͊qpqh &V=3Q?4AȈJ'))179nCkFI+*o)I[ƀ Z;5w17$g\&Nt*̀Θ>E.?>}DvNʠBQ @h$菚qu|N ZUvX$uR#Ahظˠ "0JK\ІPB=c:i}KK8_!҇ N8Qb- I_C{(<X\XX2!""NŚQbMϼ.kWw\% C6|ܹS\R'On5]rss R$({ hkn]!k$ɽXÀ:N6N KW7Ā0>T5qs/yf|X,OOϫE0!~*, ;`@@hx L! _ҵ_%pqℐޞzu0MioIjeDa 02G(|CTTDKk[SSfsL?@iq1\#a 0)M4W^8b\n.2v:t%M:ĩ66qݷLTz@f 3]R^;E֦'z+Bͼ0&`t+60`H8)`ꮏ n 1 m<.3 M vs9y: r  f@so# `DHrȎu u$7\10ɀΤQ0̛_^ݢ\ u}s\c@w$u&mi& ?(_L t(1`<7[%kL򞤮ldd*A0y΃Ѻ1(ٳkX z,ݢʗax:휜%T1}r='oXϤyA5es >UX}"˿^}ҩlhK}_J)a hc0}- ؕVӃ&91g(J{||+P\.b"Tv=. 9 )5ֶ̔Fݷ72ihóM uDj jα[0# $.+?;PŻOl<"-V-V_eĠ2+€Vl ˤ `*'; ܝ\8pvܹ.]YR_X}w̓Uؕ͏Wu 'P2wb;fLfB W===93<_+=oV﵇gc.]2G ]@ M)>U_<8dā 닼pk&͚xSǨļIeQp?x2>a8¢㈤| ߠq<ozRa:@qzܔ}vdhLUƚhOO@}ο6C }zJ0<00+#,]lu)ȂV ŨRJѻ7z׵vJ۔Pֵu_L^)k'j^9 @\,͹) J6o`KP+ap, 0]Ɣ`\$Vd 'U-Mڮ=1% N+mLp^3?_H%1Uz0ؿoAFFFzzГxȺ6% D]𒅄困JX@o0K4VAn_9Le}nʸgɖ]+/+1׍2VӗƀcY3(ԋW7\h ~Jݣ `޳8Tr37-= |P`I0±3Cjq%MHh.#Z~y*('@cEf6?yyjvSZRk @և>Ɔ` "Tj7wC^oW(Js.\ |%DN!B~t胓'Ox,Xut{tt4IDҀj7SVB]oc&QT(R/S`F0.//KO"=wU6a['M 0)u e r G  ~2DT3y! Lȑ~5ǮW^"|iCR5uJMPM'n(: 5̇Ç]#yOXN63Ǥ/NDRcV,ݻ_x}, TmTsIdC<.kgC{dz0V濽sH+E"IA=8uɄnn|@L\\77ؾmkkpC@gOv*࿡+&;ܛ '!<= n pɁ\yC*B&4MO7E 6u5Xq.Zc>HѾn!w3@k(1~8Ktx53 0׭ې@yV?){ ޙsuE>ü V^z$kڶ2ijUKKt_xlF4Kİa3.E \C!CAh'd]$./n4tEE)^$R hkz]V/j^̘ Xà,63,crx<e.D!,r+ȼT^4$V-[^ܴ) ¸t5$г(FPi3jzo SZEuפr,k4(`,LDˆ&@MK%2ŵzpmF2gB.^iBt ȿ fn9\ J35ȝ4Knn. NfOiG{ҿ)ikI.7c0P# W\]5}l.Pȱώh͖=K}cK4Q?5Of0N{(`^@;v({{ _t-h"53gN;88#v#XT:kc|gvIf9E/**>7K4KPPpyGPP cG Q Y )S^*0V2 5\G:a\vi(i8?)7hR00E}L'0JL%c ]\o^ߠ@PcyߠOBܳT(=P`QBLc>F=}m0;6 bP@ vL8 `1JeQf`=0[vb<`28% Jn\+Ld|Cx6ăj0Q|x)ޮIlLjm5&x}`Pif+T+(>HmtJU@G2`Y $~ز\]T/Z4/P=o uԠфo11?LJA-(qTFmO$0AU0A4Yb]Mx?:19! MQlx O2,`aB/0U (ߒʀ C@(N ^<:w Z`a1 .iWW[&300Vy* oXxK-y!}\xu00񲯵2J~ii7 %0dpiW}'<6Rs>5m!;, f"–>Ջ0/΀mpuEui(bz :P>ǖ`K5`-9#FKa`փ5+9}2N|]4!k`Y +`q `],2 A*my`3+7oaF\=a7*~DaG{zeX-%ק4ڦ$A[X``5 `׈WYX ùsglm&L#]v64 T?-j"B?|[ku_~y%K>wcݛ3uƦƚB"|\ .Y-[DIENDB`n1M&Jok;PNG  IHDR DysRGBPLTElf;73?=3DwU]jh]jf]{njnnwf3wΙӻΪѪӳӳճ׻׻ٻۻAgVvv~~杏w BP5(»B7o" ރl5MV$c.\ɖ*I ,TQx=F\{0 a>a[ lmCuPX[ٖEҀjN (|p°+K< i@(p\P,XhO5> pC á(j$ҞjT `(\Fax"J  ѠGaKrD9>P(pN9P8|u!˗pxK~:sFAC(| GOEf!j ?+p8~K [(LbGO_Pƀj (r vbpL/Q(9``(/LCBu'3<{. j ;Q CS(TcPHT6SC ax.,(<6t ^<d(< .~Kg6@A}3Vgo|TM(hGcfjI5YCY@ՏBꩬV (R9 iC!'?5IoR: /_,Srh$b"Иe"_ obX(QV$ 4 j _u1 &PNG@{P" ׮]+j ] -DB X -( f C'YtƋ(` B( lNf f$Zj: )vd0t66nl!lN7UCak}H@5.j q>$ 7& QWa2(LtTQP,p; 'ۓ]u( rQ.z+PRTQxHBj & ;;jd]x2PьoP}KK@|{ST =!o~}"P$H:!]KBϨS7P BBuZ ]Qbtīj# B7) np)Xe 9c ? $#RJV8 #к|}(Wl]4[n zPg` aP(( &0L|Я 6 &غ6"0woA'ۼC"%p4]xw) k-v33CI`Mmic[ KJH &6}o ;;;{NS0@A8_(.6shhUv_LxYSBPA+A??&Yx/oE/& CPxUSKW( :WDB\-a /^P3h< g%"=}ް\ CхUvrH]"fRh00Wx\FACaUV|UU5(Hϝ WNNAeೇ8*=dZ@ [|M4W0H" 4}^a +WcWّ>0_ž# uς( +Wy2ӯb3< A$ORPB`v#8 7{N8VWh#Kp? "fPP3y|F >CqXNSveJ>MB?YnDsRFaӳCz NJ gi}ݢ(i{Φg C  $ַ( cBZ>{ǯCz;?N6L:KA MaA#ᮗ`oQ@Qc`[&OAX ?kp)-H@PKFy #GC,uQ@*Cat |sCEaa=NG~.PJ@a`g^7=Rt4w;w +as.h"҂;wT`FKMDU> s1r_`X!­MD d &A y04|**[B? ~',Xr 4x|ށ`!(X=]H`9zbh yi2p?N yt@!w a}ZBG b>_I8<GaUri3>2o+ a+D<@8XlOd9Ha0B& G,x[~Ru;B֔h ~_2PPB|Pc F~0~lS95kgoZ$Aaf$ @(h %pK% 5pg%_P8MEp з*!d_&qh `D`o허P cm(зC%! u,.ZB 87:^X #%tP@K/, 0,zP@Kƣs.9ZB;6ZQ~Qz¢,yKK9&^}sX EXkWxr uNǃPquwXBQ\;ڜZ,mt8x]lnE jÆAQfsPm0 ܺ?%@#X.3..@I - ag{* ["(BWBޡP[Q P4n b4 zB(A^̵XPpKP j\Q4zB]5mq)9V~:<=Gv m4ߋ$ !~@{Rm:}N9{dč$?V @k'CG&䩆E($ C-s~0Zh:DH)HBIK׉~۩ѷ+$q2| _0%0QpNJ (tĸ~(ZF 3YB:wh$b6)!8 :l{[z6&A[%-9ϛ!V` ɓc E 9H 5qa R]2ɼJMSF0A 3G T’ ηg)!oc3^, fcs-IG!Nǵq8cwXh(W:@B.΂¬=H/B+@xVy(w:3̾懍+ bY+Hs3,Qp幠 {l P  ѣNr9HЀ 5|JV p29ǁI-+MxHg |Q4& p`ЀFfCr:X~x!0 %kEsCA !@ j1@zBX+!2ACKyG/9XWв! 3P}Z"L!+)j%WP0`\AKғSmΗNSՍ+Q,%h nE^:V(6 KpP`.AKX)40 (D—Ȼ ^ZXʲP ,X-: , %Y,ء@fc4X BX jm#%MA. +'DIo'`(G~$2 hwP՜m晴=P`Rj`lHf!Ny)˙:oWEBqp#46R,2" `uo{g*,dhL yli,lQtYH]r-u#0qf yH] G(E!L )ԥ h HAk- r/6a8#G: G"d lm`gsb!"+{)2;bwN']wR_C0 `qg1x52\w08I.Zς(ӑGG?(0`-W\ߏggImfAW+$F%I(]Cr pj[g|6*͙ECY Ix#P x1@Wh y$1oUf TXE#Q`(\?(! P_2I2 HcH VG(<&Qep %Da/Ђh# LSH څp# ACៀpYy O+[X $i}[Is|{% PN|P 3xv,1xB+_rZmPr~sN%,}; \ 5 3SOgϿX ~- y.C)fTjKe^RG͸G_C(<TՉڤ's妒pЉ I p; 6 PKUi! ?2_FL-J5D>>_ *ۖ֊4 o) B> [Qh-PU>įH+A=%M!B VP(|'/Yߦ(F1g -B!.`P(|՟Qٿ !hkb5,\!$4N )a0<=6k(h#%\{A!"}03qQ ,jMzNcp:C#IFaX@ pA!< 2Q>ރJH:jQ=i1@-=!!Ox16cUD&9 lDr|dq( pLG!}Fjd]! bp%m?+W }sDHjׅGբ {ubVq+BEp@At$r{Ry(q!>[P,u35="52Wt=}oj R vT({ܦ(2.dW {Fxf{S+0QYRFd% #rIq: qsIqj0(0WbxN$Nքp'5xPf+Ј 𿠃P cS!dC1@ݧַ8 +C. 85gb rM}Ǭi=Wp DB|]} m+T  \amԳ窅 ڇF .GXQXA?s]a@YH">q Q±\@˴=Hr% E[I)Ƅ,S +! ^(hB@fwm j3BC"O j-lA+T"De>AD!P$XYF{T\!Qh3u!8`I/Dž|\& @Z!Fa 2r̚@փ |_~PBPg"SDFP, ЮXucD+7 Y`͑+ݖE,@CI–MiKR&RXh |Bt~TK,W0a5vu505`@^B Py&a?hKܥV8-$}\a=Оl lyS3\!P&J!Dj12(y8(l'BK js/,Vq R((,PXr`'!x'd5'O$$>xqv)&_VV!2|70p\?|P8o[h@!\Wd"=#1X`$GrYBŦ0WfpGXB vU ՟lB*Yeg;?F0E)krDi $ 0pha$槲+k|v) iKeFDoaB! 0HfsQ(WrpS[;UBK}琀MWG āB[PYh V[Pw Io)T}D.>vE~ 0\HKh)8QF3J4:3AW:\ #9!&{5|Q:ʪ/{08\M4@qL%V0^k 6[ b/Y(HGXhHs5a.\DoD~a x 2Ah qHGFa% |^* E|W4 A l}NCWPG8TS;&!f`F`{{`Ly/;w4Io suCa @@+gHC( ߁ ( $Gy': GBܫ|h@;ViȈ6:INBr,6i -4 r+x8pvlO+lx5/5 ,ӣz|JSܾT>&*%o\$1]5[P00\O!Uv?90g DtƜ#D$^lcc2\W008Kx.<h,|AD F"A*w5'pMJH\#vʯ$̍0@3䷍hi 0XwP0}TxB*+A D`u{ԓ;#5C$~Y姫"a6! 8QߑxHGh_ 2|M(dxs@2wq$̝nA>b쎠zQ{F_JyYʳeO"H9l$(AB CBPX͗ Qlu\(db@$ [zghz;ef0|KP&[B,8VpF˽% kql³e,Zn&Kyx`wBH%R,l_:B,șSvQI(>RƠ Df9ҧ Ƃ]%VIBW04`.+\  ! 3OoaIi+YHB6#a\_ɗ0A0?J-G1PLKp~ A#`XqJA()h0N RVFbAKKfjZ Cz l#Th+P]Г @v/Q {3F V FjPg$(,*~ l[h'EQQ<#yj>ʬaShZ7luz=F,lU0hk sRA:B.[zpΩR GB[sd(2BwQ\(!IQQWHU(||>Σǎy6[ |WJ) ( F!D ,!u 'j `Z/= (T38R fR8uOem< ,6*B$z(4 A,5o[ jPU|B[Q4:&iӵ-k}-Cޣ ؞qB 5ؼ[Pb|N$ Qh d]A(Bu̙(dd^a*qQOB^na GʹGBXh" me%QlQF[)# h Q`{LaX =By ӐUX`] jg! c!+b,- r(@u|b7&XBciR_ߥ (Tr f:b^0[Qz,?V wqJBi*_É( +Au8Q~53,+VN,H*g" B')_`E sp^!,4+V`/txzvpq"sDm;V V z uA=0q@,jagP6|lWm}l!jNߋ-!vq6I@a{}:Y sj74?IN |fPPu0Wz'Я9(\vWHbXr~#P`,$:G#)L0R@-e FbD`uԢi7Ⱦz\G@PgA2  mj bLz@>j yD 6B5eH2 xQm 㽚ӈXxQMCfW1>@BD$rUT'xEJegAJ,1EgHD,Is~j9K yQ@uY~($XOq A3'Ƌ(TQO# :Bb&B.[ zw)  (+$ذBu(8vxQAa(NVD( `-%SYQ0H-Du4:&iӵy( BP( BP( BP( BPfR- IENDB`naLSPNG  IHDR0 pHYs+tIME<>"tEXtCreation Time20-Aug-2006 14:21:18` $tEXtSoftwareMATLAB, The Mathworks, Inc.RPLTEzDsIDATx{8`@*>rFF zLz MUH000000! n0?>N)cdJaL`g 5oUF\aw>tB>2A=aGt# aPc9 aa8:3t8nta0[1[MF!c0c(Q Lhj=|! 0 ]Qk=&G\\ C]c0!"Fzs' ];(b7 % + 1raR#̮\`{,LwtOz0ӎb3kA3J4Kra66CDQJF#xx*9ia.m&y@Q.vG19)%5.}J4mXratr*0ͩœhLR} yτ9Ѡra-<M,11ZY>oY8xg7k:0L{bÜh֒MF]. 3~<5:F )\c1FU. koI娤 {bt^ #/S8%ER.cR>^4XY {~/f<0z,]ʅ9YbaEeR.ޥ>F-ǢOa}b]b"70k #^zy`J6k3z)Ff"kwYkӠҡ\yAa`U 2ޞWLJǨ]Ct 3i6kMe<-h.0p\ʁtr);i&8Ӊ'G٥F8 .70\{.%绵eK+@oe?^Ynm 5FKL={LEI%gTI.p,6{,&'[/l Yl ,= 6w0H/`'Ysg;Kgm0`Ab3ks1($ j6,#X00.Xa5aCYH;ܦEH-fmrIY'KDrIzB,P%h_0YBVL^g,E˵^!L$|-i\X½Om`,`{,9u c #X0]hQrq)+`sZ)07s;c8nKab +W4@kf5s¬I.,c%IrCtõh') GazQ{nH.Iٰjza q%]a hLR,`4 & ơ4 Y$b[13сYtR0X?L`Rd2o] dIVcf3'Ik-sĎP0tZKR^,64JT *z&N^0/pfcW̒}$֦L-0 a(Lo/Jzh$iw6 efNҡ`rp1a fIRcC aN_,Ca$eY0JkDL Fwpb~SNX0IgQB`2 > 1^7ׂц1NRfJ8˨`IwLeI2bdsֆJP09zwQIʷ`f\0` ̔W9Ln.~`:IdfLpή`c]~+`Iʼ`n+'|bټl_cnYaU35̔ {XJ" ƭbdkmEZ#fW0Yɩb0Xiҵ6$q<;/iGI*`O !I% 8L) 3$i`u~pqa1I% OR! SS00Ccu SIڎy@´`,/`2(Lr fHR9Sw% $L`w1[d3$VL`XZoP00e LE L`pi/V0E4n% fHB fHRaS ~)& Y Żi [-P~]TN@`$ya̿ ƹbۅW066Q4LAA1ȂՎ0ẽcL7AT-Lj`w6Ai&hK,5 QU&(1&h@ =V'f`MP,&(1EJ`AIENDB`FD P߶\\y*4|JFIFddDuckydAdobedC     u!"1A2# QBa$3Rqb%C&4r 5'S6DTsEF7Gc(UVWdte)8fu*9:HIJXYZghijvwxyzm!1"AQ2aqB#Rb3 $Cr4%ScD&5T6Ed' sFtUeuV7)(GWf8vgwHXhx9IYiy*:JZjz ?}qTE-MYR#-KS,$3RUE VֳvUEGTʲК#8uyuZh!Vc`e /IdU}gzF@'PI?`=rK><-U5\m--D5,T$$u2@YKMWGRl 0 ㏡G[*Tִ=sxhTzL0)62J9ERDP>3Ֆ7ufPHE S*@Hz?[GLJX)ii(熪ij)Y`5 22 *XЃB:)V* ΔEKG>9d=B5٥|G1@> W  zuƮZJHeD zx#i&ideH%% ϽD4d:R*䞳~y}?=׭|ƓNLEF#H,>CJusĤ %iQ_E~:ꪪ5xiyiIHѪ+*"YAtE4PI8 ABB`Y-Q?=cxey9_ ʧ1E8F:7#WhGT~`ꩠ))kj9dDzh)ՈifZJYd*!#f})`H*~BTd:RjGbE\ooz@bxS$. hu)%<)e?GuwG8!>jgR=u~{ߺ^׽u~{ߺ^׽u~{ߺ^&f\4rrU$1S3ZG#J,~hytaRH%؄- *䪆#Ja1ӊ"(S:<:+s4}hST2Tw獵B:PMGI8VH]gHEcYn1lF]O:H>'dP֢:JėJg?<yɵp#{WYh{wO%d}ϰJaȴՙ)3̂.øfKxKS*TR Q7pz"~65BCiNqٽ:oe>/y4_گx_AOL{-Lnz?҇y4Vwk“'P~I$ȼA1^YZ(Y1oQlf|[}4:k[jVPXP-GC\z'E|z^}ѭWzx;6ixTE]Z-ۈoCI O08qE-诂]xm+2jkABTi$֠)ܭw:mCCϡ= J:X_6on_d Xh^:M_mZxf[.:Uucxc모h֚uSåj6%Diñ6z\]jo&z|fŴzR2乯Â@L}^{fÚM^c3Ѷ&=E=M,NĔҬ4DbUCUzw4}M({:-Tk^pC}pzKS`-6N됬 _[KBS+Ut M2TӬo`b홊Q0+iR.X^$)#Zkk%V0<34)\~@~m]׽uo/gfE=}(yDzֿZta;$`<):ǧw5n+HIQQYU;T,5,IC,Ya6%n._UqCԝ}M(DX( 0u/ ۅ^Xre]=Ԕ"IC;J`>6?G)2+S$gTfA fcRn.پ덼`Sqrrh8@J\ٌUch2|>UB_r߀;z?~|]S{_U?Q_YיnM INIw}<5zc.,Һ,ƳJmŰ%Ѵo$}MMnE j( kRT3]Qc=;xgˮ='=%fjzC6iw[^#KrPRGiGK/YZhű +e~uV1A6?ƍN :b s?**#4ߦſ6o[Iy)9{ɢ{V=pULvwv~STs Դux wiEco8hҦU#颥[$(MI|F!n5J;ڴR.,1ѤUL!Acs"+~UA c˥bUYϬzu{{^׺u{{^׺u{\BI%M )&9(1SdGd93|-yYB)yAarv$Y# 1z"3UX H]_{KEvdݱtMqҝ9IkU<UK2TzD3 H;}ZXjP-fR4$)%KRz'97Xo{Ocov.3X OH76_0ld{*kRRЗdLA,[y[imA+Apbs""Fe9k};ee H?_!N uR??#kcɌܔN.KKeSÍo}YS.Vu#A S[$`RnZ7 $dY"G5r"AP+ƒMVN$|WG;;1ؽEvm6x;r3832QIUN\<$9̛__X-פ2]+]54pw{n.ĒsZqEz~O;>-x{~N~.fnݭ_v'`UNq5tyk_(kU4;L$R(#JQP ﹓ti٪eB0YJMF:_ݕ O7cV|z,_z ~&UG˰7ۢRa Ige6";Y4jih~ts];$|ft b)gjfVy>?X}mah3;6Xjiq}e5Ykj5M; coywܹ=-nF$jJ1ka(^d I臖us[U;Qs50+N93_9:*t6{ybeܛ{ԙ\F&2U3:91>SI{yܹ_6،nAFMFE'n{HIl@5@E+}[L7=eEV[jo( 팠IyzA⁁8&Vz~?C?Ο>ğ,?i__ЯƗ'OmS{{jFW[ n˩>:SM3AI4M~]/O##}:cjnmx1 rgʔY嫞YihXIXZTiUr/n{%qṅCI4pG;IMpOSv,۔fDPqF SIoRqJW߰Gn{`uMԻ*2Ul7|.O9QZN. d欫vU332{Sm% O"D( $J5R<@Ŝ=Ou{3‰Hb$@pN~"~]+>~m_ݍma\m];eb Vo=QԛwoSRS54mEȎ=|[|6igh56ҀA+ J ҴPN\7pjA ~MLCbq WO݄KlfڽA.G0nzjJ˜.R摖HgXK>l;+6xQ"j 5G^"|ͻ&Z M ֣GϡӦ?Mt鏎{W`v﮵),;>kni=u4ԴF]e|9k=9S"y4De-  'u V++=`o9:M|uVxTt*Nn|n>ŧ9ߙ.}%d;JM&ȩ۲Uҷ9q5aOFRm6Xm7hjmGM$٥0CIdN,V15 iTfI=_g7;֦%]Yx ic+k˾FJyUIA,oK'h;emKqob0|HmrP)f$E 68^Sh#5(C%h0KWM4pQOYJQv=NevrosTiW 5} ZYX3M<^w{>6m̑.c:G{a5 '!=d?v@uh(DžM8mcv޿cs+%u'T_%;^z Ŏ{Ʀ)j:i1QG_]ZV΋ JyMO8heT0msȜql[B H\ZuRs$(=XH `$Hb*/}kCҶRTF厲Z׺u{{^׺u{{^׺u9ޣ!ⰴ0KYRcb(YdXa8⍙A?n"Y8A$TE,"F I#>`-?Z>qYm~|Vh3öwZ͹LXi!<~mh肭`ޞZDn n 49ϢnmijVD'xo G~o@>*}2}"GVL 3W#K#h"95&S'KD iKPkBVliTw 8Sυxp;%v | eL'Jf'iDK't?[oe9 7q*\9Y_?)S?wduojM1݋K b|nJV۔&Vդ&&:jt8B^"9we a:!DX/F(=k6oH*8kʜ*)~9 W6^^Ru*=Eo ]ol\l@& +\"bt.F,'1@6\x:5P7/ VXi^K }9r{Mٝ)+϶(0\N?rm#JicJjWH2y罸k搶\K!H+_AgJo60hFV>gU{_'!.A^_ nq_$e*.*X\<Ԃ*H='.sڮ8X (^T4H$TVpµV*+_ʽr$?I8m4+kcܸS;ۡۦod11䨱* lLdSLlWyܠ#YP0]A*#Tc0 _sй \Xz[Q +*v<- juºeC%&/V%-_bm=ƘMǞbw5EOP@uVQMQqS)Y%xYHN0E3O?sA_b߿?~ށ~؝S{w]χEꝵD=ߛߵ[l uDub!_'S\ fVj\-$Z]G:R  ~|?>YoOwgktRo7̜> 1>}O24m>.gXꌽ]H$yο1Otb}":ӥvNN>J9<܇xda>4u6OoM5d3t~̞Y{Cu J#KGlD 9,W_J|/]'>#TCEgn٨6cN٪suNwem ybbrFcy|n #dyFh:Ojikk]O;!`}_O~Mum۵; n 7Fy>'jC OW+PT{[D:RfV0?mWCO6gľ 7T۟nn},s>a ]E;uE?!+m]868iAW ҠUHt5>ӘCC(kmީ鏁=e)fv{νmi$autt!\[?xҕWބFڽytꜞݫIֽIٸ;be:rn1WU9ֿͤ\Lk~T?j1FI](ڒv'In(<4TAоp?.lrnԽq-Mp+IYA|y]jKIMq9.u݂N4U>9kgJK:Y.EQS>b:&0*G_pW}_HP|uxYv-wȏݑ7-CquW:Zʔ2VQc8*^WΔb|3Zn=[(>K^#eG[Po ?;>faёn"lt,%uEnUD3q^YY_*_c'ܽshz$ó>3nl{}ns2zuwengZzxd>YFYnkƲŚlYH|^]֛4>돜=GUH2-vM)K_AhidKJ}(=5a뽅l&϶jmvr[^;X[]$bL+ KMOܘnstE]"55oȕ$z :feSz$.P<.}'7WacknOlrYܮw m簟wwa㫕 9 Z/#D]ۭQv^ntocvRj4Jw? yQdIJ(2Ԥ9~}^e[S{h^l&֪ms=umf1Umg1\&\5D%-%N+S7&ݟ8wȏ} = 䖗;󰶌(,OGPm Y79wN2&,YgQ?G/7:hd7?~u'dlzʣVd;7-KwߍK򯷨r[S9/huS0TGb#I:׺J_#supoπumCf`v'y|>Ҹ+f:H7OvÕ)5&+ڬFkvj/o_mnufؽiu{ZFXrqn:꿊ΰʂ{u`Y?K%{ MEd%ThJzgHee*C+SMH@pA WIZ$fXQ.^unܝ-ݻnne:g*}9y}N.$T2*嬌E\0>oGwsڗ'œ,.("mf"4KO?ܟ%Uώt}'pt*.ת쭥'?ިvgIdn|WTRyٷ-]vZܰc#Ԗ5j@iba tAS.꧟K?2KOGGĊ'4]*ҚuhN騧Z sc)5쾪x&/,j_4ԫr@<@BU`qà|35_7S]S>S)멲UP|[I붝DUCVA)>Hji4hT{|)VqfY%q.uŪ#BIW_'oMаߌ Rn+ԛtjeC(_6o%SVGO knWw\O+f4ځmXzԐďO ܩܸ꺼ve4lUUA֜;P H|?'3zsc>&i{ר,E>oum7Z=2;%Q]ZRJ5swtwFYw]S[1}OSlLu QC_RdJK˹K&Y1v$Dm+> ,QBJ RҢxÝQ۟#/nt͗:vo ݝ?l33x,5I!i}}6bM+/ŘʬxGuKi)TQFn| ?ĦPU+_A/m$ESN*W(WOd_|T4Hڇ]W;1Mh:&J꺨Q۾%Cymx 665CkM$n,BElKq!C0H!>Cf h~,Rɗퟏ9Yi>54*8ki%m,}cb9 I]w"M,Ir.ƥ-&rxž `c""+*>*p9uϡ~|>Lnw0rڼWըRڵWQPO<4ӡ)@8><uOW~8(jp&f<] 5[;-fe2kjf8Fln!Uٝ9  KiUMFJZ tQ"R4*kO|z7wR`-|w{ғw; ]y5&ކ-߸lsN附V/훖l}:I\]۝|~0I!Ns֧?iK+%]rY?U1[%}W_>5;9z9f08ij<Q@ᡌ3n[33LY3#L(T\ a)?Ek > vyJ[zev~;:۩g{whIW|{`b+L8)*1.N pT}[l {kVeY7,ZY9,I=5h@|@͙s˵f&['lxu*nw+%U]uL򖚢gu`6QOvVslm֘>؝&tgn h0sl+# T2un-? (w??eEu{; kn5Nnl7!Vi5(Бp wIfOe.&uvʮ}6~/-z,LM=\ 4n~׺Yzg66SN[ . w⠏r[{niYa1ta)3WKc7|Ns7 ,mY mt2sEuk7bq$aKUaIZ4h)V5U'>A#vNRs[+^;jD}kxY3Qv,M,OQR$A1\</Wi"h iRI@(梒><`r,e6j}]+Ӻk 읩5QÓT[IEQ<5B̨}!Zͼ)klDAAfTuW@ T09w1$ :sG;j?w=AV-?{ _FOw슬C#jLwuO61:a`'=/`AQ9-bjZib88]s̐,km *=,( FK.jH$Xƕ:4=j Wˬ qu}f.f|GqMUޝQ=~,$3cTQVIҴS$2ּ:ZG2TUб HhFcF@#MbjZP3ƤPҹU _❵[_Z֞q;_꾊~ײ2p?nvn%yyUjZp8IjyegFQ_6X< *i ڒ$FÁ;on,,2Y{UREh EĊJ.dw$gI'{_Nblj0?`/u{{^׺u{{^׺u{{^׺u{{^׺u{{^׺u{{^׺u{{^׺u{{^׺u{{^׺u{{^׺un'Q³f#*^[PNG  IHDR0 pHYs+tIME 9 n"tEXtCreation Time20-Aug-2006 14:12:55/n$tEXtSoftwareMATLAB, The Mathworks, Inc.RPLTEzD 9IDATxm8 `zz| P (no U$ $%ٲkwv Lq1S)A:/4 3¼?ż&`\EaD'&vݭ]8ǥs `B 3,CF%7M~\\s%dcq}m.`YfxyRf_ES474j| 0.aqyS`Rb24,>X0AX\Ai oPK8(M>C .(-hd%# f3,(0 X)BL(.p̧|L .Y P@92g)`Lhe* m)]p,9aH  _A= re)39yxNӒyXKsǖ-X21d0?O\hYjqeK̂\2o¸gYjq`綵D-\0mEc\,òcQCgN `].3 _6Yaɬ0e+ SE0r?w:E,i3~}"sdb\ґca:K>a,%b0"YHbv10Յ]w_\k2pOaLLpK~x^-V'_[e9 \ƻ@ g c,9L^Ѝ۽# L˙(`s}T;@K,aY|K uM^<5S&2z@,$pA. .?9907f#`.IcʳS`%NL `zP$<ޙ$qcOP0}Dp,A<|g 40|.,8_*>|8Kg9P@?>@ L˞ӳLKѳd &ɅkI⧬%p~zxi@ۃŒh]X2NMcK΅%I}iQ,Ya|\0eQp a2NkgbjWRӣ].Y%.C"x#dKQ10|Ɯr cKdmy`#%z`DcIv2*-3^yq 0]Y%̲%̲%ʐr)fr)¿n8o'ʹ.HOI/.i@0ku9tcV{\`RRSڅS܅&50]xa c7S LP-сr)ς Z`jqI[ Uw@).ԩ#2|2TKY~v $azT`jra;F ױ<0\`C4&8>f.,($A/0k!=,02]g#ԅ(0R]`@xo{P-ׅPH]at9u!F :߸\ĻPJ w ~/݅f[6N'D\طsmD\CR ]EA~`H[{BDFx/qąp7nŅZBƅ&*0">&&05Jʅ[D`rYrܒfYHݝ Ru.40[z߾BǪ.]00d>] Kv20.1q~G Lԅ`ɼju/ZrxS X L.HV\a]}vA1VB ]hz%󿁩߅tEUK.0?C5(ƆKl.oFKg4oC` 3u,gtlַn9]NCI- aۯI$_.0dqK ˱ r?fib Las?_`aM.X7W傂1L0\$0v`C"\ Ka#c TC9.%0\|IoQ0\r#E 0#Q0\dt#EH02]HHۭ#ץ,`0]J\O] v)swRzOa0(#l/Y̓E00 Y{lC~ȑV5aE4'77bfQCۅ2{gm/?qzP0R7jnVi}zH0 0 0 0 0 06d#ԧ<-%G))&#FK1MaFaFaFaFaFa>X$4qIENDB`n Pr;vkmhPNG  IHDR0 pHYs+tIME v4a"tEXtCreation Time20-Aug-2006 14:15:12Hp$tEXtSoftwareMATLAB, The Mathworks, Inc.RPLTElx IDATxi:`~9*|٣)aԫI%&atI&[00000aaaaaa98LlB%0vx3 3 fkRf L 00] {rۦm6$L۞BD_R.LImk¤jШt[Qa`:>tZ oO`:`_Z0ҩh)z _a5BK~/0 (: 0 4@J&itMKc)YJ*1b,Lo?j c`bO^3bS(4t2c#,˜H9 @:O'0va at\WjP $SXզN*0 Nc (SǭLs[ 9Gt QZ|eVmjɄ!V?کYc2`4i f"GF~TJ+z^_)Siه4ΖH1YP~0TdT#'SX:U#4sL0"T%L0$a؃ZnzaIfm.-c14C8sK?YYF>8F&1tQT_r8`3L sP_Vd 7vpcI~i laBntƯ4,e[>`b:鮟 E*3s L(?3|81 ٕ ^X8eZ!wC(_ Cm/,.3 C&0"`b_|aUd0ʼn 7h.D"0r\K!"FFJʼn02B30n PFqS02a4el(X7#s215%0J2"SFT fz\92`boFW#+SKW FP5 n;DTA֘ME+5DLRhS02Ha2Cn'QȤdta?tҒa-0{m^6)Evx.g}(ZQX*uMzc)MA\ΪTd2ڌYKz][`jLx=^OR>AI&~[(45V)Xisjeႆ!U1+sҁWf2T:32TE>)s!K()Z2Cܵt*M2qB* 7A-3q h< 4eM䏮!_T$d`02$Wl+Id1LĸOQ2hvMd|Xo֐ FU2K&l0?K8 rd! Yٟ>QG c9 DD`F5f#FkBLQtY`/&#FkߢY\MQTT|sadZ=&eeDLq- `daeh[6f`̄Qi%a@){;(+酖y%+R)I&@tdJn J2!#AE&?d&e5:*2W!]5bR):oJ&3dt2Ii@2Wo"y&2itMYW F焌LNUWkʨ*xȨ+/뗦5CFwZ `hXA&1m-.[c`X2 {mv6kr[l07 g^D 3F̴0/C ɈF $d0D5.#N-bH^ΦjucT/3m\siYM;x^tx\NkSA#{ RL2Zɝ6_<%41[9$7A"F2꯵#N-M*B.Ztv,*4m}02#k]"4Lw-Nskf&W@F CFeɼU:9h&o02' Fw-N-FO.h&ޠ oSo@дU0ʼ_f44-e_Lҁ%S~1ÔN4ɄiVm C2$gzÓN/!caɼtz_Kͳ,jS&LFG-tL0t!-ӐL!`HөO&=LSWvN`!Z!|[=L2Lg)~n`۷`Y./_pɧO0ty3xgal+9Ô4[mH=6qZ؄=0A` Q,gox!6ȓn@N${r"SLLRLXRc 0lA;EbF oT|E` L4r0As&tAH` o@N_|waaaf4qqqq0a&N'wc aaaaaGV6zpIENDB`nj*f78+7YPNG  IHDR!%C. pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14" IDATx EO>*8'1D'. }{JiN1APtM&؇(3j1'VIl.z6"P1eW`+ ԡ!nvy ^%:$h⌤ 5iXS[?o9-=q\fnK Bi& }򗿬j\X,vuu-[y=Wy+pt=ya^QYzS7;qdkUjm~D(ֲ&WfJ?E1nR5R ㉇'K81>di8pbb,]K֭[Ǫ>ˏ9뮻nd$YC+K܎ch !#$Ed I@&;`̆ahqoƺQw4HVN", r VLQD4L;4 9XM4"+V*MCS`K%LJګJ~ϘDDK/ԑՁg֭z+uiw ˗/C<ƵgTlziiɤ?qq0qLWRy2gB6)? mJR82im0Xu;ad5ЪQ36Pb#(Cg$1WB-fԠQkt3j|80l?ّ={f#R)ʎ-[B@! "[hV#]*io~[G>wE/^KMYѻBFɄfbƳfHd؛;Cs iEz݈|ӱDhZ!Z超4Aѵ 4׈!cUke1#ʹ?3۩U[rb_Mo~"/s}s=_=}@(\dN=0Y{.SN9oqEW$”0L21U S""xIH@Hh%hAw¨l*Ci2$r+4a؊R mDJf^v8oq4=߼+$tvB8I1J$8-[aCZYhԒ%K#;8 ^aƕj\eZXpr4ZՁ3ҎWi=V`f)0ڹjVssnUM7Ah+zί5L;I$g{&ˬɱƌK) S,ˬ>ḏY#Bh]=@`kkz7bPq%$*zݣt<@k25kּ5 Jڶm8CC֣>vZSF'{Q:гiE˃OuYs| z^9w틌:ZQJOeؚqb?[xC] $(m6};4mX:Ů̅H^^Nm7ϖ/i^OWjU@*alB1F5 bpDcT޿s~ir޴fټLW CֈE4M¬^-^cO># o.53fe8jZ@ lh$MT ume{I'%}QlӦM !+f8HQGOȟ,r'SK(޶!65 qcyZT89Xz[v|4|Fdz~q௬f8 @ SjT zzeײ6:ƫdD'!QOIJIBXǵF DYo9[::A(]^k{D^- Q [}VZKNkث]^o<ċF=B-A~Q06eH %ѵ< њ=׿%<;;)C,EJ]dBNRE,?TOBN jݥ7cgt lUDUU7=h!a+"վuKfӅS{Mևwa ՒP⩄T`Vf584|ktSzϦgܾ|ip˻>N=`gM/qW?yq[dcç\dk犛s#|K{{Ӹl]+{*; #YJCf"*b;CMxCB4xx4TV:SoT3v j:\H׫/=kQ{pp-G]-C\=7`4s}uu r?R*62a"k| o~HsgN-'"y;wl7o>t,v ЂzKMV2Ek,6 ԵEo+?ś>+:>f7(h)0R \U8O5̰̄¢+W:0-PDPxz0N4!0QzW Mnݐyٚžs9 !qiEu2=f&7l,~\ճsjՎmǷZ Sr??I8aczf3t)G_k[#zݍ]vqz#J9֩Yzں/:ba-!MlPt KxaҁM[[U|]Gw3uyandziD!16I@D@*1tAe@R K (L2udIVPtlJLfFxqӭ0.ۆxb(+g杛Yy0aczZʶ2I 8yOFj`-XL6Kg>[ArƯmyXa{7 4rtUkM#msvefYv VusЄgYύMk:{h˛=cܗN8ͻۑϔ V~̂E@=AAO9Kb(22$bNDM`:x T4OeTk@)$LtzdvNPdqh6oz%u8@\SF#.i"hjK"ka 5lO\eVic '@3|Gd5yy2'/|oӹ?9blbPȄhrF- Ur@DsKV,=|xYg!pS,2ʤ)$:+T> bbӯs=Ty\/vR=}Tw6]H¼Qxџܡ|W[ñ_&:),Nv81X$t&jM1qk2SaRx++bˉmi(~w`\L Sju@b:Yz:2:,ujҞ߾okm'G~[_Ԛzz怟ZX0bwx' Z4,ع+?s )2ʰrg'B(WpJDN?t Q>Yd*o2lV=juGAe9*Ϛ#φc&o=eW \G;yյq࿍z*\^N$9@^KHr [N$,SMMVmzjjilrtdo87FsxooAZ09CȂ +@=\[o=!]x|4lK/7['l^ -L \ˠU!F! O|@GP9O&M]9wP(21RlshBq Ch~ ٬jۮϖVǀQɈ dt W3ŒFt)6ܒ &H{Ҏf\=|)ίh'҄Egdv·2*^ԁ2y͡|W.1.OT'qljmQὍɱ/shb=jy%DͭNlqwt}g] H1v*~DDXDb:%0ƕb.Qc=+_Jbg>8||d E].?]徝tG '?9Zwe7VO)0M 惖Yle5kNX8K҉M_"S|F;#EssUKġ3>d^njJQ6ZkX L2:,WUl4BMJH|w %_`c%g?˙/bHw"CdOta'^=Xujp^3qi8n. zO2TtOņ!VܑhM)Xʼ鵄!ƼcU%Feq|Q,V=|AKC4'&✐k@~N:6uO7SgRz,;o{T w)J(]o9Cي:H&I|b+bp rR!Ǔ蒉.K/-!r.0K@@G0b"K.x OҦ"+ c#5uwKnAr傛 Dobγ#u$،B.u4CҞ|Mt*d,5%J.fbu]H|[O0?\[֯_;Lc.҅:IaCҗ]!9䔾 $_ Q [jX81M9MM=QSB>P/P/`T>Q e;^UŋAE+Oe,= "K],xʙZPvNxJҎ1$$PR@yL&߾L_ dJ]TjB#B=n!z%W?P՝Ҩlw%iX-L)] ŷHEhCC(uA TeeWdRog&*DNW[PT$BZ2|, e`UjlJHXfޚBdt6aqD` ZVdV zt#͸3CnKj%1TIٹdQR{lQPu}6׷ GKNdDtfK%Ҿ-p;+K ɷo2<QhoJ k6~rNybr U(b_󥕦ؾلUfHeskx AKWG3W_}5V;Q.."^LKqFYtȄo2y*d1}B2ش l8,Z@Ő#Z .6"uz a5q l\<䆯2d*AC-aOi8ۮmi6al92]# ¥؜w"-2r?td2";te?BoMY?FI?(&D'g!F&Dh'9cdu\NNms_?)6bӈtN5JM<cBPJLHO\:rk5TūQ1Y?^Y!(' ͐fgNױY̶BD H`V)|h^f$eBIENDB`n'UF5*$ n&bPNG  IHDRv pHYs/tEXtSoftwareAFPL Ghostscript 8.13luIDATxK P.r Qr,LTLUuZ`/`(&.%?H]""&sM~*a.WH,Eibr^.k!A9[n\BbY*. O^'")")⧒X^2& DV ښlk.n*I9%.lR`&l" ^KzGj=ɳ!z_AI)[(Cg"X;Cʹ N~spIIÉsj.o44yNS[A/:8,]Pfa@yk[vhRD7̖Wt{;0tyMݛOm;~#[_$WRh_yY~]VQIENDB`nܮ$mf\yPNG  IHDRB۰ pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14"IDATx]v8PWDQeUY~p-} |swཿ{A}}~^%RjsU5 0{0ُ^寽޿ ?9þy[%y{*y}_ᕩ[f9g1_ ?>gæZfW|r qΆaZ 1X @s6k,Tǘeв8{^U3*ؕMZ,,^jF_qWc1׎z%z1K[4-=\_GSF+WW욦3hzw"9;u]BKӼ/Y?C8ozk`QNfW4Ե]UR?t/ރ[:u^=F}XQRi|#̕ӎ.fŹߒW^\.mNav>0^l:o戽ut4,]m]ivHu&q4ׁ̑rիr+ + r|k1C~mz{=X|X]i+Z}M}fޒΖhM0X|XHiÙl_Th :9:5X|"W<.^\.u~d-Lo[*E)=nk=U_.z8c\+B@~/2/Oҧ)lװs\%,pρDLhë,?svui ,XUjr.WW7~&yb5E8sܨ4S?X|$g/ƴ/޾^6yYNEA ,#00><_a-#h//bKhKA\B/|8νR:, 5N9]Q!ʮse`8B~NC`8UOη{}̲v׏O>ۃ?8/f=LFlr3#V[61n.<[&XW4c`A;f(1 LXlNߜe~4n9/oFtlG!W9s3rU_!+#%鬜\S<_[k:SNj窾Ox3Ol\q"9 5yRDW(ذ k[ÞDG^E}kLq yf+g|d31J_a^*-)/MV=XVƒ~8(H|+--T5B# PWˮR1iFn9˖vAe_@R\_ѫYWX~fe9WEX|+*;=F>՝m5-/ɮ5޼FkkZZOA"}h(x)_\^fC9u7+@#=NQspLULz};iׁjwkS1^1yXKs,4+ 7`;J'>|1؃1Lw, Y$Afs62X]ۜxn]IsaBo A#9c]m@(/޾vS /NFKg B151_ zG8+*WTdG5-nj ^z}++z D]>ذS?;Pq 9Ƚ͌F()zs-K֑S".LKXmٵN^_.uYHڔ11 2 LlisLx޳u:bwOaEڠJ@RͲvc! fvn`ߖHsbwn_8nOنI_# U>9;QNW7i{sDC6a7´W!7հ5ubt\=FzW RGk)<6 تyb$t_)2= ߻Kޮм9iUxK|lpҪ sA8Wg /,< |W5f1_H`9n~t͛&jE (6/>\ͬ>e'4[J-J=@`czO$zȳΩMDzj Wj s+GS)ĵ${97 yhC|4C/JN`kI+Äe9I`>_P.Fˮ;_,É2[g@K94hrZuBi]fSD68{?+ͩ#o5zyx|DF5|.a[M%mݻUZ(wԻ?U,;>>}0wqo{'?'$Q{N1.;?{vmڢv4kl-(Oz%j9ǧxk,^1i<>~֑3dzvz *Sm|؃r«'ڏ7 Ix!ЌK>WzJMz^r1R$ ?/^8Zo~j\V-I5Dp"sjs8DC4mOjGPV'YleO qmz]u zLx;nfQGk>MqBFϟ% ]o"m  i\-c힋jոUA jb]Pj)E3 "]'[؏atjqz|z֛2F#j]eכx6ϬkKzͭfX5YT3oN6ӵ p 1nz`W])W=&M;%ήW,pzح:A{ӆ _ ^(/\tQĴT+<ԦǘnUsBhCQj_gہ}Cuˬ+Vn+o;/fiWuh-,ʂ Ҝ)FA񶈨rjɮlƾju@N]U cLCs*{:[ީ-{-Uˮ mkm^d=ưM3e;%o.bUp.d^2մ?T\|EۢWR A3A6vX7۪eWfQ㹡 cfMkU6 pVM&L]f=rm~YScL{_gܣl s~˴ysV-2Wޫ՟7u=ϟ?gGd##P[; ٵPۍW4op~]zYqhsYfl՛.ic}}0]RA݅;|]Z\zŽo11ثכ_/EJ[.zFI]pY_p7 S⽯Zހ 6>D5`AnZ9XWa?)nn:N-yR*وL~."ieWÃdM$ISo)@ %'_]ס^~}n+񶈎  niÃ:=gioXӿa7j<0gϫyNXJG77!cxO| ]*A7"OW2N P] Ҏ;}= c*|!dɀ;%Ymԝ<|*-ͱbusjWoHBɇ ]iڇŹNTb_mT}=u_颿xfB}s+観. `:/xw&i"Yw=@)e&z-zuMWTo|}0TK]j3,2⟃dW}N|krj$ۖS29 Lj㽯JݼO9HA3&[$hh?l(bUo(Mm]T@qWc;,U~}~xTVɐo90z"\f2^٠{*`n}27\clk]vӕ7zޒy|pl͞u<geƍYJ%bw ^I@U{0'E'izK,'ӛ=x@w1K9sˮWXG 2 IrboX 篐,PA^/|@3S;?Gz?yT_Vg<%y4D=ۮoq dWN#,sѽxE~qZgہJ;y`^eW4=Tc3\|"LuR`^Ru{&U^o$إNt\ Yjy2:ʮd5ܯXɲgꩯ;wm?3K!l Cumzߩ}}~=}yf4{XCv 9J9oQ'@}:窹Ns (J :u/@ g}>jdb0+nS7Poy1O|M2 Y+,㍯+ W*~`0I|z5*o6h N|+:XG6z< <^z'}}~-LSW-gW7Vrb?l Ob{ȮowoIhL崟ԴAͮ hL|];UDe0Y_-iMp|+,j?-k08VT;GH Զj $`!gƞͮTWX u0e}1'|q׻,LmSU{ ,rŕWZ!`f!*D0N].b@Ek蛳07-( @+GT;qM$X ֒9*ۘ6<0X  KH $ *fSƀ(bg4;X+%mu_WT,0 R|m3f^_{cWv~zؿm4+p-}bS ls*Ֆ#خ_=ENjZvxp Q@VQc˸Y_ pz+ٳWy i*(ZBz"KS=*O_ zSp+fX>ٽ5?|[Eb[fW3|.P蒐X~u0rJSMjܯ: LJ|N]c{Vqo]c*0g< uuE@49]B "`+@S6 T"{Fii[BW+K2ʮ@U+Ae,ʮ@mwHKP}sWc+@(F/O u6]gB|ɜ%w͛-90(9]}\tM@@?gsTW^LW X `+?F2IENDB`nQ3N8Lkؾ>PNG  IHDRS pHYs.#.#x?vtEXtSoftwareAFPL Ghostscript 8.14"IDATx @7[DIU UtkFDL,|$qίvL[+&FLbuE`<fE=~%q"sz;&WH\ x΃z>'`-{{!!+3zB)wXNeYS$|4GzKW읨ޛ0ÛJ=^S=6ճ1dm#[缽K vv@q"~qqZ$,u92>@؂+$u;RuIyBF'=N]G= y2R%{+HXy/uk˔txd꬯_W05cdl]~#|9/qL:c;0%604H+oyKUF@5LRbu@U;“M>ENY<7nOt&] ]CQbD3?v@Xe S 8p3`zM.XLc(ѵ_0ʠ(PWľiB%2R_h\}<dCĭ߻A* Z 1c֟Pb}ˁCc=z2n W$=@a`{ck)O( s>_oWx eVqw{.G*QFc\2RVO~3,dkB/LI+ u{m.0Y+9\vi lw:=01,#҅q1GSMM'Q!/ ,!<m^[:(%6tL) l}ךWυ#PZ>AR=~[?9Hd8YRH*_m+;(ت݋GtR  l'5ߍS: CS`t_EL#A#3`L_1L5t9 LcNɠ֥nXR=q?^JF(=vF;=O};/ 2zFz<aBLYo*O~U&.|>Sq +u1\E ފ|J~yWyJYE *ϊ<=Ze]u&Lf11|byMT B?5]0Ia3IENDB`@N(/ 0DArialgbatl?\|<% 0a 0 "DHelveticaatl?\|<% 0a 04 " DTimesicaatl?\|<% 0a 04 0DComic Sans MS\|<% 0a 0 B@DZapf Dingbats\|<% 0a 0PDCalibrigbats\|<% 0a 0"`DGenevagbats\|<% 0a 0pDOttawagbats\|<% 0a 0"DSymbolgbats\|<% 0a 0 c0. @n?" dd@  @@``    .&  &&)/ )x G    _ ]$$ %   b * p ! # b$ٮ5FE3]u7zn u=b$xa"NTyݨIl u=b$cmSұ 4eBP=b$-qqԤM1+=b$,:u^J] 3=b$H8$^3CG@=b$ono * C=b$4!+tCSD P=b$VJ^]/o]u0 _\=$=b$8sb3ef=$=b$Wh~t6 )h=b$gL,)i)Jx _q=$=$=$=b$(%QnO!|=$=b$M&Jok;1=b$LSi=$=b$_S盛HIruh\"=$=b$o' 59%~=$=R$ P߶\\y*4|D=$=b$Q³f#*^[/Q#=b$Pr;vkmh 1=$=b$j*f78+7Y,?=b$b!0nDR5h2 @=b$UF5*$ n&b/`=$=$=$=$=b$ܮ$mf\y*c=$=b$Q3N8Lkؾ>|=# l8c?`!3M@8FG+ Hʚ;y0ʚ;<4dddd= 0g4=d=d% 0hppp@ <4!d!d 0\h=g4PdPd% 0p@ pp <4BdBdf 0\9)___PPT12 %h___PPT2001D<4X@___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography 2DEFAULTFONTSIZE10.DEFAULTWIDTH4000DEFAULTHEIGHT3380___PPT10 ?  O  1LRobust Task Scheduling in Non-deterministic Heterogeneous Computing SystemsM M M qZhiao Shi Asim YarKhan, Jack Dongarra Followed by GridSolve, FT-MPI, Open MPI updates VGrADS Workshop Sept 20060r  h4    L H"General Task Scheduling Problem  Task scheduling: Allocation of tasks of a parallel program to processors to optimize certain goals, e.g. overall execution time (makespan) Application model: task graph (DAG) Node  computational task, has an execution cost Edge  dependency between tasks, data transfer Computing system model Network of processing element (processor memory, unit, communication via message-passing) Optimal task scheduling problem is NP-complete Heuristics: polynomial Some assume every task has same computation cost, (homogenous tasks), some assume arbitrary cost Some ignore communication cost$bZ/$bZ  /$  i I#Robust Static Task Scheduling  Heterogeneous and non-deterministic resources Application DAG with task execution distributions Traditional goal: Minimize the overall makespan based on expected system performance Our Goal: Find static schedules that are more robust to varying task execution time Scheduled performance should be relatively stable with respect to the expected makespan. Approach: Use  slack to absorb task execution time increases caused by uncertainties Employ genetic algorithm for optimization` Z wMZ w  \)6Robustness  definition (I)  mRelative schedule tardiness : Makespan of the schedule obtained with expected task execution time : Real makespan with realization of expected task execution time Each realization of expect values gives different schedule tardiness Robustness 1 Reflects the amount by which the realized makespan exceeds the expected makespan` Q Q   n Q&8Robustness  definition (II)  Schedule miss rate Robustness 2 A simpler definition, simply counts the number of times the realized makespan exceeds the expected makespan <%m m  R'Calculating Makespan  Given workflow, processors, and a schedule (assign tasks to processors) Adjust communication costs Makespan = longest path from source to sink  `+Defining Slack   JSlack of a task node i is defined as follow: - bottom level (longest path from exit node to node i ) - top level (longest path from entry node to node i) Average slack of a schedule Usefulness of slack in improving robustness Slack at each task of a schedule reflects the  wiggle-room that task has. Large slack means a task node can tolerate large increase of execution time without increasing the makespan L/vI/vI  b,Bi-objective optimization  (Want to optimize both makespan and robustness (as represented by slack) Turn out to be conflicting goals e-constraint Optimize one objective, subject to constraints imposed on the other objectives maximize average slack of the schedule ( ) subject to: Where HEFT is a well known, efficient algorithm for serializing a DAG and assigning a schedule Use genetic algorithm to do the optimization Fitness function is average slack For solution violating constraint, fitness is penalized by the degree of violation of the constraint H! O<_-H! N<_  -    d-Experiment settings  =Task graphs: Task number ( N ), shape parameter ( alpha ), average computation cost ( cc ), communication to computation ratio ( CCR ) Best case execution time matrix beta is generated taking into account task heterogeneity and machine heterogeneity b_ij - the best case execution time of task v_i on proc p_j Uncertainty level: degree of uncertainty of actual execution time. UL_ij- the uncertainty level of the execution time of task i on processor j The real execution time is a uniformly distributed random variable The graph has an average uncertainty level UL=C5r"4C@     e.?Bi-objective optimization improves both makespan and robustness@@ @  h/<Relaxing e improves robustness    k0Review  /Want to schedule a DAG on a set of resources Resources may be shared, or variable performance Thus,for each resource, a task has a distribution for its execution time on that resource Want a bounded makespan for the DAG Scheduled performance should be relatively stable with respect to the expected makespan. Schedule is statically generated (not modified dynamically during runtime) The schedule should, as much as possible, be able to withstand variations in the execution times of the individual tasks Optimize the slack subject to constraints on makespan`-$T-$+$ 0 S$ Conclusions  We developed an algorithm for scheduling DAG-structured applications with goals of both minimizing the makespan and maximizing the robustness. Due to conflicting of the two goals, epsilon-constraint method is used to solve the bi-objective optimization problem. Proposed two definitions of robustness Slack is an effective metric to be used to adjust the robustness. The algorithm is flexible in finding the epsilon value in certain user provided range so that the best overall performance is achieved.  L GridSolve Update  GridSolve 0.15 released 05/2006 Support for NATs and firewalls Based on GridRPC Support for batch queues Improved problem descriptions gsIDL Matlab, C, Fortran client interfaces History based execution models Scheduling using perturbation model Win32 native client Lots of work left Client bindings (Mathematica, IDL, ...) Service libraries (ScaLAPACK, ARPACK, ...) Backend resource managers (VGrADS, Condor, ...)L  t,  V        ? m1 FT-MPI Update  FT-MPI: Version 1.01 Fast, scalable fault tolerant MPI implementation System reports failures, recovers processes and messages User must recover program (via checkpoint/restart, ...) Status update In maintenance mode Improvements in stability, performance:;;  w3Open MPI Update  Stable MPI-2 library  version 1.1 A mixture of ideas coming from previous MPI libraries (FT-MPI, LA-MPI, LAM, PACX-MPI) Team: Indiana U, U Tennessee, LANL, HLRS, U Houston, Cisco, Voltaire, Mellanox, Sun, Sandia, Myricom, IBM Multiple simultaneous transports (ethernet, myrinet, ...) Resource managers (ssh, BProc, PBS, XGrid, Slurm, ...) Production quality, thread safe, high performance Tuned collective operations Fault tolerance Involuntary coordinated checkpointing Application unaware that it was checkpointed (by SC06) FT-MPI technologies New FT framework in progress#Xj&7#Xj&  7      (             M!The End   / NOU V!X#Y$]%_'a(c)f*g+i,l.o/x1  ` 33` Sf3f` 33g` f` www3PP` ZXdbmo` \ғq>9y`Ӣ` 3f3ff` 3f3FKf` hk]wwwfܹ` ff>>\`Y{ff` R>&- {p_/̴>?" dU@ <6?" Z2@%  d $ << f MMM n?" dd@   @@``PR      ` p>>v 0 (     `xaxa1 ?"#  X Click to edit Master title style!! [  Zxaxa1 ?"  [Click to edit Master text styles Second level Third level Fourth level Fifth level !   \ `  s *<61?"Qy\  c &A"fc`  s *<61?"`  c gֳgֳA! 8c?"  0 "d  \*0  B  s *޽h ? lb *VGrADS_template_Aug06  y@|(  | |  `xaxa1 ?"3p  X Click to edit Master title style!!   | Zlxaxa1 ?" `    [#Click to edit Master subtitle style$$ ` | s *<61?"Qy` | s *<61?"` | <48c?"Gx F0 \ | c &A"fc | c gֳgֳA! 8c?"B | s *޽h ? lb/ 0  ""`$(  $ $ Nd?Rectangle 4097P    X C   u $ H d?Rectangle 2"mgPK!'[Content_Types].xmltMn@ zU2 LXs851I3 w{m J , a_""; ϏpO$&YZ=(eLY9iDلZZY4Nx jv~6(8`Ja𨹩rReN><Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!%?. drs/shapexml.xmlVQo0~`u`Ҫ=Ix&!qrgw}wf_HZytLDZӬ5:P)HĘ+o?*c[2ܫl\ƹ2nm ] L>u4 aBΠk+~ݲ\Z%ay:=Q$`-vQu5(_ T" 8H QrܡD@pr딇@LvtW,c7sXK]F0A[% (ȳ4 5kvs눺Scࡌ~y%yKP`͑3TFuKw@ż9 !;a956"{W* -T @z>QfNisbe!'X-( 4]lI,Lr3Mj3n`+b,*'CePjLc;cju66nd?j몚J}Ǘ xyx]N!%Kq&S?z~φI5Zɬ5NZAw͢^:ÛqN|] a<۪ 3A\U(V4`kKl+ h/pPՎew*2?qVȜ6VL:5%)]ReKO:rԆ}X^"˰B z A@_ Ւ\pf =3$J>O,qnP_lgj|~OoH6GaPK!drs/downrev.xmlD]K0Cx\b [W EЮ M1eݿ9ͨ$ngЕn(&ӜIm`XA/B&)І`S} X6N0E`0!sc:=]̊m+(| x\j/|qNKC{sYdSOO xPK-!'[Content_Types].xmlPK-!1_a _rels/.relsPK-!%?. drs/shapexml.xmlPK-!Kdrs/downrev.xmlPK\   d p* C    $ T1?Rectangle 4099  d $ Hd?Rectangle 4"PK!'[Content_Types].xmltMn@ zU2 LXs851I3 w{m J , a_""; ϏpO$&YZ=(eLY9iDلZZY4Nx jv~6(8`Ja𨹩rReN><Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!V{M-drs/shapexml.xmlXO0~iBg:Nձ3H&>H5w}w߹r]JZ4>P"Y)}z=RbSZ*M/G_\TU6RpJ(=ՕP˵)fUFXs৔Y3Q EGpZMA߯&YJ(V!G&3b6 %37. Tbt8ڐDnq llSf C!uJްyMXM <'s0;PMJ{]`X;A pZc!4uwB@\JAYYVc낳։7Q%n} P%B uO!6[J7ZQЎWyф|wimoG.H@ r3"ʀPs $#M1=_u6VfMrܕJ$R5j?&p|챓jZqV|Y186i"om ^vvVHE4^>t&Bϱx, ຬˢ?)dRN1'Jf^Lj}J04©(Y o FHcXqێb+~3f,p'FeN*\- a'V"Ç-0 uXPV"ρ-X5HxL#3}+Չt`6`nDKxk#gV(}Iv$Α808*D%ّ$v+PI.ϹHy<$Ir$|$ԛ$ ?,YrdAW ~>FPK!#drs/downrev.xmlD]K0CxodKWq̺ls+S&͛4[-pS Kj F\Ky0<Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!E drs/shapexml.xmlVn6?^[cBE#0)*VM*IN>rȡ(ion æ[`ڲ*3M-μd3>bR8ʈRYAjJG'H±ݎ<ۼ&bTk۾%<,;,8G,Ӌ9ȏ0YJXت%Ca!FW㍩BwT>{ nϜYn*9'-Dr~S>0W#+YUh,P\o$EzNb=>U@Ԓ]aG{+Ld4O0?vȋl: ` 7vF)p0zpcÊ[Z`,!5׉@6Ћ?F]=a7x8~b.-xڦ"UpޠTԪl&/|4.Ye7rX?E0p [XTe߷MklRAWɥ}$MIMU>(p{o^ڠܱANĕ:jqKN&(9 S;S:\Y6JgTS5^TL ǏQ'x6&TZ=R~.z㐔½q7h ~|`{ʍJ]Rq ѿ즯eL>'ẫPK!1drs/downrev.xmlDQK0Co.ˆL2|Ve-6IMpzRкh3uedk>I8/AX./"f;_.RW5u*C5V:161UAш7 V q[?|(DQ'ί; '?@0l62>Q$;%a[Yr`@W?PK-!'[Content_Types].xmlPK-!1_a _rels/.relsPK-!E drs/shapexml.xmlPK-!1bdrs/downrev.xmlPKs_  d p* C   < $ 0޽h ?"PK!'[Content_Types].xmltMn@ zU2 LXs851I3 w{m J , a_""; ϏpO$&YZ=(eLY9iDلZZY4Nx jv~6(8`Ja𨹩rReN><Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!7drs/shapexml.xmlAn1 EHC=$S$Fa #'J7Ă;_zϿމ+<Ь1r "d.0icvlSNV=PK!Bdrs/downrev.xmlLAK@B2ovb4R<(m{fM4;viwK|0Ǜh;1hZng͵S- #uv,'WxM4"epIBc#іNoPy:l]-^4F׿<h +{3R^ODcLmvn*NBZgNIJ?PK-!'[Content_Types].xmlPK-!1_a _rels/.relsPK-!7drs/shapexml.xmlPK-!Bdrs/downrev.xmlPK I}OPM80___PPT10.U  PK![Content_Types].xmlj0Eжr(΢Iw},-j4 wP-t#bΙ{UTU^hd}㨫)*1P' ^W0)T9<l#$yi};~@(Hu* Dנz/0ǰ $ X3aZ,D0j~3߶b~i>3\`?/[G\!-Rk.sԻ..a濭?PK!֧6 _rels/.relsj0 }Q%v/C/}(h"O = C?hv=Ʌ%[xp{۵_Pѣ<1H0ORBdJE4b$q_6LR7`0̞O,En7Lib/SeеPK!TGtheme/theme/themeManager.xml A 0@ѽDд;wC3dRP=Y~qv.5dp4(% 8AF(fa aL[9</R YX[ b]RKֳxt"^PK!,߼etheme/theme/theme1.xmlYKo7 Hȁ%Kq81,%E.˘\ߊX@Ѵ@o=m$@/qM]ֵ!~fr~4.DG4I3nֻ\*!)A0=$J囵 aK<'|rb"Emjkz֬fp S4֤VI?`)B&Fx3 6:lh<}&f$ExLu7 S@ X=:\3")90 Acg8 +l e~]oxFkoXR%z^d%|w<Y|g ?ixJu 9dn%|@籣ELy΋b dX Lq1njN&;ʥ!- P\ur {컗Ϟ>xzӇO`8݉/>B?ţϫ?Y5P_<_~۷*O\Dl3 kN&'3X k)x' \+xG@}^%b {]OR9qQ ׵,YW 3wQ><f9FZEO>Ù1ɈB?$bR |]zV.ɘNhZLڥ)JA6{wP*wȑB1a2^3*1N7J ODŽq4Usn qunc'Vq~OpWaG4K\B}{~ٹCm{*-D K^Ni"\L3u<"⻃g>x_ށ޷7xkyj]Yϱ is)elN!M,aÈ0癓 b$dFIbYP\vxhؕ=pSg9vbs,55š R0U5R+KkLM.ՄNA܁ Hn¥[^d#b2 iK+ v`\#Gz;iei8JVR /ij279YA:@!λ"xJspԽf1\Jب0M/ypcqNye R`0Hdd_kC̼&mZRL$Tr|%186\Z+8(KN[1l}\Vx 7uu.Ҽ9ڀ)[VѼ%J])B#L2eem/\ poFz}Cz\C=rDž@p_1Y.,hF%} 3Q&h.$+)L8 ˼u01u\/+{cw}Ɣ46¯9%#cdhn PK! ѐ'theme/theme/_rels/themeManager.xml.relsM 0wooӺ&݈Э5 6?$Q ,.aic21h:qm@RN;d`o7gK(M&$R(.1r'JЊT8V"AȻHu}|$b{P8g/]QAsم(#L[PK-![Content_Types].xmlPK-!֧6 +_rels/.relsPK-!TGtheme/theme/themeManager.xmlPK-!,߼etheme/theme/theme1.xmlPK-! ѐ'j theme/theme/_rels/themeManager.xml.relsPK]e ; *& `(3(  ( ( H4?Rectangle 1P    X C    ( H̨?Rectangle 2"uoPK!'[Content_Types].xmltMn@ zU2 LXs851I3 w{m J , a_""; ϏpO$&YZ=(eLY9iDلZZY4Nx jv~6(8`Ja𨹩rReN><Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!~6 drs/shapexml.xmlVQo0~`ub$4j:ԡ#q ñ3aP>LڇM`Ůl+Jy1L\ZnqfV"JwMbgMk~kQVjSíY#PS NG*Lm* q;oEVRw>LA?R@6JLW==$fC <[ p2cCR'ɮ4롎!eΣhx#S)9$bXi<9چǧHO- Xw- b* 3%C]\/;(5C #6''Z>1"[ϫo!r+c,'Ž1>B4x<= { B j03ʓ~Jo~G,,IȔw0SِYA4gVuf ltWzn.n"`2P!']˪~Cs] ϡXn$(K=Lu[{-yJ݇Z0ܾ0SrLws[ UMN̄/_/Af PK!drs/downrev.xmlDQO0M5͵[K\i(Ŷ%&Ǔs,ףnIk r VNCyR4^,E*̀}E&TpRJ]Y+-t l~ҊarE4Z48]ԢSYʯEsEv ?E${ƧG ^ eag2A8~/C\\?l#ye90PK-!'[Content_Types].xmlPK-!1_a _rels/.relsPK-!~6 drs/shapexml.xmlPK-!Sdrs/downrev.xmlPKd    r* C     ( N?Rectangle 3_P   X C    ( Nd?Rectangle 4"|PK!'[Content_Types].xmltMn@ zU2 LXs851I3 w{m J , a_""; ϏpO$&YZ=(eLY9iDلZZY4Nx jv~6(8`Ja𨹩rReN><Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!򛵝C drs/shapexml.xmlVMo1W|Ȳ|VY" %= ^/l[K U{Jɡx͛\]JA\BɔmJd*+&֐cAf )^?~S+MRtkmDa[^Px+]ŭDK "ۃPH:Fgrڭ~IOcs6`b4POdczPn5q$ [m}]]"1ǔ~A[)B>W91pW<'.p]JR: \c% :x2/|3`q6wEj,/uxzUbS=95bCO͈`mLb >K:#]'7t^X_bc}kLyc9a%x D zSŽ_`P[ 0 %+D9{=&H7ᰐSs3Ar%_*JNlgjYKcLŖ̒=۸>MdkM=$kNL臃u:>/͛XX.$iX7"zc2gy5w͠5lޛ^v'ϝOr*\hʮ.R}-BArz\B֡Jvb+_]N]S(ZaMtWGo`(xTKTצ3]9R !Bz(2os7`b_Z k*%^%0iXB da>?[T۞vC&TIBhS}QQ~f[;ZgT+vܦ<Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!7drs/shapexml.xmlAn1 EHC=$S$Fa #'J7Ă;_zϿމ+<Ь1r "d.0icvlSNV=PK!Bdrs/downrev.xmlLAK@B2ovb4R<(m{fM4;viwK|0Ǜh;1hZng͵S- #uv,'WxM4"epIBc#іNoPy:l]-^4F׿<h +{3R^ODcLmvn*NBZgNIJ?PK-!'[Content_Types].xmlPK-!1_a _rels/.relsPK-!7drs/shapexml.xmlPK-!Bdrs/downrev.xmlPK I}OPM80___PPT10.UpRܴ PK![Content_Types].xmlj0Eжr(΢Iw},-j4 wP-t#bΙ{UTU^hd}㨫)*1P' ^W0)T9<l#$yi};~@(Hu* Dנz/0ǰ $ X3aZ,D0j~3߶b~i>3\`?/[G\!-Rk.sԻ..a濭?PK!֧6 _rels/.relsj0 }Q%v/C/}(h"O = C?hv=Ʌ%[xp{۵_Pѣ<1H0ORBdJE4b$q_6LR7`0̞O,En7Lib/SeеPK!TGtheme/theme/themeManager.xml A 0@ѽDд;wC3dRP=Y~qv.5dp4(% 8AF(fa aL[9</R YX[ b]RKֳxt"^PK!,߼etheme/theme/theme1.xmlYKo7 Hȁ%Kq81,%E.˘\ߊX@Ѵ@o=m$@/qM]ֵ!~fr~4.DG4I3nֻ\*!)A0=$J囵 aK<'|rb"Emjkz֬fp S4֤VI?`)B&Fx3 6:lh<}&f$ExLu7 S@ X=:\3")90 Acg8 +l e~]oxFkoXR%z^d%|w<Y|g ?ixJu 9dn%|@籣ELy΋b dX Lq1njN&;ʥ!- P\ur {컗Ϟ>xzӇO`8݉/>B?ţϫ?Y5P_<_~۷*O\Dl3 kN&'3X k)x' \+xG@}^%b {]OR9qQ ׵,YW 3wQ><f9FZEO>Ù1ɈB?$bR |]zV.ɘNhZLڥ)JA6{wP*wȑB1a2^3*1N7J ODŽq4Usn qunc'Vq~OpWaG4K\B}{~ٹCm{*-D K^Ni"\L3u<"⻃g>x_ށ޷7xkyj]Yϱ is)elN!M,aÈ0癓 b$dFIbYP\vxhؕ=pSg9vbs,55š R0U5R+KkLM.ՄNA܁ Hn¥[^d#b2 iK+ v`\#Gz;iei8JVR /ij279YA:@!λ"xJspԽf1\Jب0M/ypcqNye R`0Hdd_kC̼&mZRL$Tr|%186\Z+8(KN[1l}\Vx 7uu.Ҽ9ڀ)[VѼ%J])B#L2eem/\ poFz}Cz\C=rDž@p_1Y.,hF%} 3Q&h.$+)L8 ˼u01u\/+{cw}Ɣ46¯9%#cdhn PK! ѐ'theme/theme/_rels/themeManager.xml.relsM 0wooӺ&݈Э5 6?$Q ,.aic21h:qm@RN;d`o7gK(M&$R(.1r'JЊT8V"AȻHu}|$b{P8g/]QAsم(#L[PK-![Content_Types].xmlPK-!֧6 +_rels/.relsPK-!TGtheme/theme/themeManager.xmlPK-!,߼etheme/theme/theme1.xmlPK-! ѐ'j theme/theme/_rels/themeManager.xml.relsPK]e ;  {s0  (       h0e0enn9    ?|?Shape 1"PK!'[Content_Types].xmltMn@ zU2 LXs851I3 w{m J , a_""; ϏpO$&YZ=(eLY9iDلZZY4Nx jv~6(8`Ja𨹩rReN><Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!QxG drs/shapexml.xmlVMo8@H;@"m#0{,FeH-I9N~}I)ur*Er7C_~8VV s&TRmlʙurZ+?,xwYmͰVys׳NTdZ( m*rm6 JEWQæ^oeae!g*Q-؀Z % Xlo[30rCHG4[{&stRp`<&|4flam\nR0..ixG}2^cd8MG3b E!2_8a61w2c^Zw20pFr 9əBއ6PfMaK8<੃F5*V)~ yei@3 F#`4 10kSQ*îA(jr+3/][6NjR.i=Jĺ<* VY`Cb3bA"|墸ts1^J]gZeXD ̣֍%;R[4u2ٓJ:sQZ(Hiĕ(^džֱwY.VĝT ~ ԦټAσ[(ŕ &ظ[3hX47`vTeTdpξl:H34+:SC0}@߆@=Ly,ħ0L Y"F"yizfm宥 loT718c,s?HD[Qaܪq[xÅ->J,vbPM|ӧ>&Y_74gj\>8}=.ygCh7A >ZlhD@?PK-!'[Content_Types].xmlPK-!1_a _rels/.relsPK-!QxG drs/shapexml.xmlPK-!)drs/downrev.xmlPKd     p0e0e|?Shape 2"PK!'[Content_Types].xmltMn@ zU2 LXs851I3 w{m J , a_""; ϏpO$&YZ=(eLY9iDلZZY4Nx jv~6(8`Ja𨹩rReN><Ir f=7_PK!1_a _rels/.relsj0 ѽqCNo^K [ILcX&m߾0XFo;>0xM e`|X}đ I`߽N4aG2$RKIZ)4(M9`ctB{m:f@`3n|O,ܗr޾jxR0T ,0@}WBLǬ5vPK!Qdrs/shapexml.xmlWn0?$H-`F Jb'-Ň^= 7oojҺ9:7EoSg ]eL6&q ^&KsR½5Еcio{Nj/NƓ7'\pM$Ĕ%htGd< D"7nr@1LѤj۵Y6423d+MPUcXfIUQ 5 rQY [WK%\VPksU)oQUA/w%Mld⮕,K-ALwtL' y3֚ZI{)?8'~ʾaґFb.@~Q ?ņk8R8N@(ϡzD}hsOM%CGhs :4g~Eš TQyVO19שn!P{a`wy:FPK!^drs/downrev.xmlD]K0Cx\ ?V 'mMwڤmhA~prP hQDpKPsl l)7ދJڧȠ  XcW0Jѭ:I: ZX>÷[ۣˋHCz6{Ynݵ g0RON= PK-!'[Content_Types].xmlPK-!1_a _rels/.relsPK-!Qdrs/shapexml.xmlPK-!^drs/downrev.xmlPK ` %   H  0޽h ? I}OPMy___PPT10Y+D=' {= @B +" +PK!B?&EFR\Ǡ~of=t0] r`6* PK!/7[Content_Types].xml=k0BZlJ)34֏!|E퓐.!=;J@!yiS/+\h=V_'20BϘY6elLvN i1&CU=CoQ}Bw$ǰ$T X ab H"tww$\+nb$^S$r֊l..y1_I^d53˝PK!֧6 _rels/.relsj0 }Q%v/C/}(h"O = C?hv=Ʌ%[xp{۵_Pѣ<1H0ORBdJE4b$q_6LR7`0̞O,En7Lib/SeеPK!TGtheme/theme/themeManager.xml A 0@ѽDд;wC3dRP=Y~qv.5dp4(% 8AF(fa aL[9</R YX[ b]RKֳxt"^PK-!/7[Content_Types].xmlPK-!֧6 "_rels/.relsPK-!TG theme/theme/themeManager.xmlPKy  p0(  x  c $\Wd#  d x  c $XdI d H  0޽h ? I}OPMy___PPT10Y+D=' {= @B +   0(  x  c $y#   x  c $@z  H  0޽h ? I}OPM___PPT10i.»PՖ+D=' = @B +w& % %% %(  x  c $]d#  d x  c $md d $  dA'txp_figB0  SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown} \delta_i(s) = \frac{\max(0,M_i(s)-M_0(s))}{M_0(s)} % \begin{cases} % 0 & \text{if $M(S)-M_0(S)\le 0$} \\ % \frac{M(S)-M_0(S)}{M_0(S)} & \text{if $M(S)-M_0(S)>0$} %\end{cases} } $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m0ORIGWIDTH301.8756PICTUREFILESIZE6540  dA)txp_figJ & SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}M_0(s)} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m.ORIGWIDTH 56.8756PICTUREFILESIZE1403  dAtxp_fig0$ SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}M_i(s)} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m,ORIGWIDTH 52.756PICTUREFILESIZE1372  ^Atxp_fig`  SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}i^{th}} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False0TRANSPARENT False,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m&ORIGWIDTH244PICTUREFILESIZE814  dAtxp_fig   .& SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}R_1(s)=\frac{1}{E[\delta_i(s)]}} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m0ORIGWIDTH163.8756PICTUREFILESIZE3411   dA txp_fig"`U SOURCEz\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}s} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m,ORIGWIDTH 7.8754PICTUREFILESIZE388H  0޽h ? I}OPM___PPT10i.ĻpW+D=' = @B +k # z(  ~  s *d#  d ~  s *hdN d   dAtxp_figh` TL SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}n}: \text{number of realizations} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m(ORIGWIDTH2586PICTUREFILESIZE2921:  ^Atxp_figz$ SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}n'}:\text{number of realizations that has }{\color{myblown}M(s)-M_0(s)>0} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False0TRANSPARENT False,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m.ORIGWIDTH 538.756PICTUREFILESIZE6199  dA txp_figUH@ SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}\alpha(s)= \frac{n'}{n}} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m.ORIGWIDTH 93.8756PICTUREFILESIZE2333  dAtxp_fig  $ $ SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}R_2(s)=\frac{1}{\alpha(s)}} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m0ORIGWIDTH129.8756PICTUREFILESIZE2911H  0޽h ? I}OPM___PPT10i.ĻpW+D=' = @B +W $ QQ`aM(     BCDEFcN7*%DP i i vK#W0|<|<RPU^anX$}OVlB(  -=|Yy4%X0h9SiPoGn@pslOG(>.*<!Un]I?!8@>&#?0BO`n~xpkfe_ZRM6!  $(-/4Yf8,@_SZv+rljeI{4W',2:mCPT._nrxv|reC@ )&M_xw}6@#<:4)  vzh}[kBc@1I>hzc@                                                                 `"`L,$D 0~  s *d#  d  ` c $dD 0<<$ 0 d )2  # ld?jJ?"6?@`NNN?Nls O1 C  )2  # ltd?jJ?"6?@`NNN?N4 O2 C  )2  # l|d?jJ?"6?@`NNN?N O3 C  )2  # ld?jJ?"6?@`NNN?N  O5 C  )2  # lPd?jJ?"6?@`NNN?N O6 C  )2   # ld?jJ?"6?@`NNN?N= O7 C  )2   # ld?jJ?"6?@`NNN?ND O4 C  )2   # ld?jJ?"6?@`NNN?Nwe O8 C  L  @ c $\L   c $\RL @ c $|*L @ c $vgL  c $v*L  c $TL @ c $jTL  c $*L @ c $]  0dI@0  g(a)4 2 C   <2  # ld?jJ?"6?@`NNN?N  bP10 CK  <2  # ld?jJ?"6?@`NNN?N  bP20 CK  <2  # lh?jJ?"6?@`NNN?N s bP30 CK  <2  # lh?jJ?"6?@`NNN?N s bP40 CK  F  S 00F  S zzF @ S 5 6 F  S F  S |F @ S |   0 h0  g(b)4 2 C    !  `?jJ?"6?@`NNN?N  "  `?jJ?"6?@`NNN?N  # <d"`< l bP10 CK   $ <dh"`-<  bP20 CK  ! %  fh?>?"6?@`NNN?N  M1 A  ! &  fh?>?"6?@`NNN?N  M2 A  jB ' BD> P ( <h"`Q  E O0 C   ) < h"`qE O2 C   * <#h"` E O4 C   + <&h"`7E O6 C   , <l*h"`P? O8 C  XB - 0DjJA ! .  f-h?>?"6?@`NNN?N| M3 A  ! /  f\0h?>?"6?@`NNN?N M5 A   0 03hP g(c)4 2 C    1  `?jJ?"6?@`NNN?N2  2 <89h"`o0  bP30 CK  ! 3  f4>h?>?"6?@`NNN?N M6 A  ! 4  f$Bh?>?"6?@`NNN?N M4 A   5  `?jJ?"6?@`NNN?NR  6 <TEh"`0  bP40 CK  ! 7  f,Ih?>?"6?@`NNN?N7 M7 A  ! 8  fxLh?>?"6?@`NNN?N@b M8 A   9 0,Oh R10  2&M   : 0$Shs Q1  2&M   ; 0Vh  Q8  2&M   < 0Yh6V Q4  2&M   = 0T]h Q1  2&M   > 0`h$ Q2  2&M   ? 0xchDd` Q1  2&M   @ 0fh` R1   2&M   A 0ih  Q1  2&M  Kz @ z  B lm,$D 012 C # llh?jJ?"6?@`NNN?Nh z@ O1 C  12 D # llph?jJ?"6?@`NNN?N0  ` O2 C  12 E # lsh?jJ?"6?@`NNN?N`Z O3 C  12 F # lwh?jJ?"6?@`NNN?N qr O5 C  12 G # lxh?jJ?"6?@`NNN?Ntf O6 C  12 H # l0~h?jJ?"6?@`NNN?N a O7 C  12 I # lh?jJ?"6?@`NNN?N@  r O4 C  12 J # lh?jJ?"6?@`NNN?Ns l'2  O8 C  T KB c $ ) T L c $)T MB c $ IJ T NB c $CT O c $FCT P c $W[!T QB c $Gl!T R c $ [ T SB c $  T 0dh    g(d)4 2 C   Z U s *w V 0Xh   Q0  2&M   W 0hp   Q0  2&M   X 0,h   Q2  2&M   Y 0Ph0 Q1  2&M   Z 0Dh  Q0  2&M   [ 0DhYy Q1  2&M   \ 0xhj* Q0  2&M   ] 0h[{ Q1  2&M   ^ 0hb> Q1  2&M   _ 0h`P S 1   2&M  H  0޽h ?              CDKCELDIMEFNEGOFHPGHQIJRHJSFGU I}OPM___PPT10.˻0B&6+5{D~' {= @B D9' = @BA?%,( < +O%,( < +D4' =%(D' =%(D' =4@BBBB%(D' =1:Bvisible*o3>+B#style.visibility<*w%(D4' =%(D' =%(D' =4@BBBB%(D' =1:Bvisible*o3>+B#style.visibility<*u%(+7 ( 66%)5(  r  S Hh#  h r   S hi h   dAtxp_figeh PH SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} \[ {\color{myblown}{\sigma = \displaystyle\sum_{i=0}^{N-1}{\sigma_i}/N}} \] \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m(ORIGWIDTH1336PICTUREFILESIZE3294  dA txp_figa  SOURCEj\documentclass{slides}\pagestyle{empty} \usepackage[usenames]{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \definecolor{mygreen}{rgb}{0,0.4,0} \definecolor{myblue}{rgb}{0,0.33,0.6} \begin{document} $$ {\color{myblown}\sigma_i}=M_0 - {\color{mygreen}Bl_i}-{\color{myblue}Tl_i} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m0ORIGWIDTH187.8756PICTUREFILESIZE25836  dAtxp_fig; SOURCE\documentclass{slides}\pagestyle{empty} \usepackage[usenames]{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \definecolor{mygreen}{rgb}{0,0.4,0} \definecolor{myblue}{rgb}{0,0.33,0.6} \begin{document} $$ {\color{mygreen}Bl_i} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m.ORIGWIDTH 27.8754PICTUREFILESIZE8352  dAtxp_fig SOURCE\documentclass{slides}\pagestyle{empty} \usepackage[usenames]{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \definecolor{mygreen}{rgb}{0,0.4,0} \definecolor{myblue}{rgb}{0,0.33,0.6} \begin{document} $$ {\color{myblue}Tl_i} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m.ORIGWIDTH 26.8754PICTUREFILESIZE7748 l )`t    BCDEFcN7*%DP i i vK#W0|<|<RPU^anX$}OVlB(  -=|Yy4%X0h9SiPoGn@pslOG(>.*<!Un]I?!8@>&#?0BO`n~xpkfe_ZRM6!  $(-/4Yf8,@_SZv+rljeI{4W',2:mCPT._nrxv|reC@ )&M_xw}6@#<:4)  vzh}[kBc@1I>hzc@                                                                 `"`L,$D 072   3 rX?jJ?"6?@`NNN?NH  O1 C  72   3 r?jJ?"6?@`NNN?N\   O2 C  72  3 rhɟ?jJ?"6?@`NNN?N   O3 C  72  3 r̟?jJ?"6?@`NNN?N   O5 C  72  3 rП?jJ?"6?@`NNN?N   O6 C  72  3 r\ԟ?jJ?"6?@`NNN?Nz @  O7 C  72  3 r؟?jJ?"6?@`NNN?Nl   O4 C  72  3 r|˟?jJ?"6?@`NNN?N S O8 C  Z B s * ! Z  s *.  Z B s * v3 Z B s *C  Z  s *r ' Z  s *  Z B s *s F Z  s *  Z B s *9)  `  0s    6m0   Q0  2&M     6x(P +  Q0  2&M   ! 6+` /;  Q2  2&M   " 6i=0 \  Q1  2&M   # 6i(P I+  Q0  2&M   $ 6i5   Q1  2&M   % 6i   Q0  2&M   & 6|=# l8c?`!3M@8FG+ Hʚ;y0ʚ;<4dddd= 0g4=d=d% 0hppp@ <4!d!d 0\h=g4PdPd% 0p@ pp <4BdBdf 0\9)___PPT12 %h___PPT2001D<4X@___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography 2DEFAULTFONTSIZE10.DEFAULTWIDTH4000DEFAULTHEIGHT3380___PPT10 ?  O  0LRobust Task Scheduling in Non-deterministic Heterogeneous Computing SystemsM M M qZhiao Shi Asim YarKhan, Jack Dongarra Followed by GridSolve, FT-MPI, Open MPI updates VGrADS Workshop Sept 20060r  h4    L H"General Task Scheduling Problem  Task scheduling: Allocation of tasks of a parallel program to processors to optimize certain goals, e.g. overall execution time (makespan) Application model: task graph (DAG) Node  computational task, has an execution cost Edge  dependency between tasks, data transfer Computing system model Network of processing element (processor memory, unit, communication via message-passing) Optimal task scheduling problem is NP-complete Heuristics: polynomial Some assume every task h  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~      !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~ J     !"#$%&'()*+,-./0123456789:;<=>?@ABCDEHIKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz|}~Root EntrydO)pGPicturescCurrent UserDSummaryInformation(XRPowerPoint Document( $DocumentSummaryInformation8<DHelveticaatl?\|<% 0a 04 " DTimesicaatl?\|<% 0a 04 0DComic Sans MS\|<% 0a 0 B@DZapf Dingbats\|<% 0a 0PDCalibrigbats\|<% 0a 0"`DGenevagbats\|<% 0a 0pDOttawagbats\|<% 0a 0"DSymbolgbats\|<% 0a 0 c0. @n?" dd@  @@``    .&  &&)/ )x G    _ ]$$ %   b * p ! # b$ٮ5FE3]u7zn u=b$xa"NTyݨIl u=b$cmSұ 4eBP=b$-qqԤM1+=b$,:u^J] 3=b$H8$^3CG@=b$ono * C=b$4!+tCSD P=b$VJ^]/o]u0 _\=$=b$8sb3ef=$=b$Wh~t6 )h=b$gL,)i)Jx _q=$=$=$=b$(%QnO!|=$=b$M&Jok;1=b$LSi=$=b$_S盛HIruh\"=$=b$o' 59%~=$=R$ P߶\\y*4|D=$=b$Q³f#*^[/Q#=b$Pr;vkmh 1=$=b$j*f78+7Y,?=b$b!0nDR5h2 @=b$UF5*$ n&b/`=$=$=$=$=b$ܮ$mf\y*c=$=b$Q3N8Lkؾ>|=# l8c?`!3M@8FG+ Hʚ;y0ʚ;<4dddd= 0g4=d=d% 0hppp@ <4!d!d 0\h=g4PdPd% 0p@ pp <4BdBdf 0\9)___PPT12 %h___PPT2001D<4X@___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography 2DEFAULTFONTSIZE10.DEFAULTWIDTH4000DEFAULTHEIGHT3380___PPT10 ?  O  0LRobust Task Scheduling in Non-deterministic Heterogeneous Computing SystemsM M M qZhiao Shi Asim YarKhan, Jack Dongarra Followed by GridSolve, FT-MPI, Open MPI updates VGrADS Workshop Sept 20060r  h4    L H"General Task Scheduling Problem  Task scheduling: Allocation of tasks of a parallel program to processors to optimize certain goals, e.g. overall execution time (makespan) Application model: task graph (DAG) Node  computational task, has an execution cost Edge  dependency between tasks, data transfer Computing system model Network of processing element (processor memory, unit, communication via message-passing) Optimal task scheduling problem is NP-complete Heuristics: polynomial Some assume every task has same computation cost, (homogenous tasks), some assume arbitrary cost Some ignore communication cost$bZ/$bZ  /$  i I#Robust Static Task Scheduling  Heterogeneous and non-deterministic resources Application DAG with task execution distributions Traditional goal: Minimize the overall makespan based on expected system performance Our Goal: Find static schedules that are more robust to varying task execution time Scheduled performance should be relatively stable with respect to the expected makespan. Approach: Use  slack to absorb task execution time increases caused by uncertainties Employ genetic algorithm for optimization` Z wMZ w  \)6Robustness  definition (I)  fRelative schedule tardiness : Makespan of the schedule obtained with expected task execution time : Real makespan with realization of task execution time Each realization of expected values gives different schedule tardiness Robustness 1 Reflects the amount by which the realized makespan exceeds the expected makespan` Q Q   g Q&8Robustness  definition (II)  Schedule miss rate Robustness 2 A simpler definition, simply counts the number of times the realized makespan exceeds the expected makespan <%m m  R'Calculating Makespan  Given workflow, processors, and a schedule (assign tasks to processors) Adjust communication costs Makespan = longest path from source to sink  `+Defining Slack   JSlack of a task node i is defined as follow: - bottom level (longest path from exit node to node i ) - top level (longest path from entry node to node i) Average slack of a schedule Usefulness of slack in improving robustness Slack at each task of a schedule reflects the  wiggle-room that task has. Large slack means a task node can tolerate large increase of execution time without increasing the makespan L/vI/vI  b,Bi-objective optimization  (Want to optimize both makespan and robustness (as represented by slack) Turn out to be conflicting goals e-constraint Optimize one objective, subject to constraints imposed on the other objectives maximize average slack of the schedule ( ) subject to: Where HEFT is a well known, efficient algorithm for serializing a DAG and assigning a schedule Use genetic algorithm to do the optimization Fitness function is average slack For solution violating constraint, fitness is penalized by the degree of violation of the constraint H! O<_-H! N<_  -    d-Experiment settings  =Task graphs: Task number ( N ), shape parameter ( alpha ), average computation cost ( cc ), communication to computation ratio ( CCR ) Best case execution time matrix beta is generated taking into account task heterogeneity and machine heterogeneity b_ij - the best case execution time of task v_i on proc p_j Uncertainty level: degree of uncertainty of actual execution time. UL_ij- the uncertainty level of the execution time of task i on processor j The real execution time is a uniformly distributed random variable The graph has an average uncertainty level UL=C5r"4C@     e.?Bi-objective optimization improves both makespan and robustness@@ @  h/<Relaxing e improves robustness    k0Review  /Want to schedule a DAG on a set of resources Resources may be shared, or variable performance Thus,for each resource, a task has a distribution for its execution time on that resource Want a bounded makespan for the DAG Scheduled performance should be relatively stable with respect to the expected makespan. Schedule is statically generated (not modified dynamically during runtime) The schedule should, as much as possible, be able to withstand variations in the execution times of the individual tasks Optimize the slack subject to constraints on makespan`-$T-$+$ 0 S$ Conclusions  We developed an algorithm for scheduling DAG-structured applications with goals of both minimizing the makespan and maximizing the robustness. Due to conflicting of the two goals, epsilon-constraint method is used to solve the bi-objective optimization problem. Proposed two definitions of robustness Slack is an effective metric to be used to adjust the robustness. The algorithm is flexible in finding the epsilon value in certain user provided range so that the best overall performance is achieved.  L GridSolve Update  GridSolve 0.15 released 05/2006 Support for NATs and firewalls Based on GridRPC Support for batch queues Improved problem descriptions gsIDL Matlab, C, Fortran client interfaces History based execution models Scheduling using perturbation model Win32 native client Lots of work left Client bindings (Mathematica, IDL, ...) Service libraries (ScaLAPACK, ARPACK, ...) Backend resource managers (Condor, VGrADS, ...)L  t,  V        ? m1 FT-MPI Update  FT-MPI: Version 1.01 Fast, scalable fault tolerant MPI implementation System reports failures, recovers processes and messages User must recover program (via checkpoint/restart, ...) Status update In maintenance mode Improvements in stability, performance:;;  w3Open MPI Update  Stable MPI-2 library  version 1.1 A mixture of ideas coming from previous MPI libraries (FT-MPI, LA-MPI, LAM, PACX-MPI) Team: Indiana U, U Tennessee, LANL, HLRS, U Houston, Cisco, Voltaire, Mellanox, Sun, Sandia, Myricom, IBM Multiple simultaneous transports (ethernet, myrinet, ...) Resource managers (ssh, BProc, PBS, XGrid, Slurm, ...) Production quality, thread safe, high performance Tuned collective operations Fault tolerance Involuntary coordinated checkpointing Application unaware that it was checkpointed (by SC06) FT-MPI technologies New FT framework in progress#Xj&7#Xj&  7      (             M!The End   / NOU V!X#Y$]%_'a(c)f*g+i,l.o/x1w& % %% %(  x  c $]d#  d x  c $md d $  dA'txp_figB0  SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown} \delta_i(s) = \frac{\max(0,M_i(s)-M_0(s))}{M_0(s)} % \begin{cases} % 0 & \text{if $M(S)-M_0(S)\le 0$} \\ % \frac{M(S)-M_0(S)}{M_0(S)} & \text{if $M(S)-M_0(S)>0$} %\end{cases} } $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m0ORIGWIDTH301.8756PICTUREFILESIZE6540  dA)txp_figJ & SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}M_0(s)} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m.ORIGWIDTH 56.8756PICTUREFILESIZE1403  dAtxp_fig0$ SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}M_i(s)} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m,ORIGWIDTH 52.756PICTUREFILESIZE1372  ^Atxp_fig`  SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}i^{th}} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False0TRANSPARENT False,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m&ORIGWIDTH244PICTUREFILESIZE814  dAtxp_fig   .& SOURCE\documentclass{slides}\pagestyle{empty} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}R_1(s)=\frac{1}{E[\delta_i(s)]}} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m0ORIGWIDTH163.8756PICTUREFILESIZE3411   dA txp_fig"`U SOURCEz\documentclass{slides}\pagestyle{empty} \usepackage{amsmath} \usepackage{color} \definecolor{myblown}{rgb}{0.47,0.26,0.09} \begin{document} $$ {\color{myblown}s} $$ \end{document} 6EXTERNALNAMEtxp_fig$ BLEND False.TRANSPARENTTrue,KEEPFILES False.DEBUGPAUSE False*RESOLUTION300*TIMEOUT (none)&BOXWIDTH400(BOXHEIGHT338"BOXFONT10(BOXWRAP FalseL2WORKAROUNDTRANSPARENCYBUG FalseD*ALLOWFONTSUBSTITUTION False4BITMAPFORMAT png16m,ORIGWIDTH 7.8754PICTUREFILESIZE388H  0޽h ? I}OPM___PPT10i.ĻpW+D=' = @B +b   #(  r  S ,i#  i r  S i i ^ # 6A8c? H  0޽h ? lbrV L\[ "2 {9N(/ 0DArialgbatl?\|<% 0a 0 "DHelveticaatl?\|<% 0a 04 " DTimesicaatl?\|<% 0a 0 ՜.+,0     On-screen Showicl $ Arial HelveticaTimesComic Sans MSZapf DingbatsCalibriGenevaOttawaSymbolVGrADS_template_Aug06MRobust Task Scheduling in Non-deterministic Heterogeneous Computing Systems General Task Scheduling ProblemRobust Static Task SchedulingRobustness – definition (I)Robustness – definition (II)Calculating MakespanDefining Slack Bi-objective optimizationExperiment settings@Bi-objective optimization improves both makespan and robustness!Relaxing  improves robustnessReview ConclusionsGridSolve UpdateFT-MPI UpdateOpen MPI UpdateThe End  Fonts Used Design Template Slide Titles$_# Asim YarKhanAsim YarKhanas same computation cost, (homogenous tasks), some assume arbitrary cost Some ignore communication cost$bZ/$bZ  /$  i I#Robust Static Task Scheduling  Heterogeneous and non-deterministic resources Application DAG with task execution distributions Traditional goal: Minimize the overall makespan based on expected system performance Our Goal: Find static schedules that are more robust to varying task execution time Scheduled performance should be relatively stable with respect to the expected makespan. Approach: Use  slack to absorb task execution time increases caused by uncertainties Employ genetic algorithm for optimization` Z wMZ w  \)6Robustness  definition (I)  fRelative schedule tardiness : Makespan of the schedule obtained with expected task execution time : Real makespan with realization of task execution time Each realization of expected values gives different schedule tardiness Robustness 1 Reflects the amount by which the realized makespan exceeds the expected makespan` Q Q   g Q&8Robustness  definition (II)  Schedule miss rate Robustness 2 A simpler definition, simply counts the number of times the realized makespan exceeds the expected makespan <%m m  R'Calculating Makespan  Given workflow, processors, and a schedule (assign tasks to processors) Adjust communication costs Makespan = longest path from source to sink  `+Defining Slack   JSlack of a task node i is defined as follow: - bottom level (longest path from exit node to node i ) - top level (longest path from entry node to node i) Average slack of a schedule Usefulness of slack in improving robustness Slack at each task of a schedule reflects the  wiggle-room that task has. Large slack means a task node can tolerate large increase of execution time without increasing the makespan L/vI/vI  b,Bi-objective optimization  (Want to optimize both makespan and robustness (as represented by slack) Turn out to be conflicting goals e-constraint Optimize one objective, subject to constraints imposed on the other objectives maximize average slack of the schedule ( ) subject to: Where HEFT is a well known, efficient algorithm for serializing a DAG and assigning a schedule Use genetic algorithm to do the optimization Fitness function is average slack For solution violating constraint, fitness is penalized by the degree of violation of the constraint H! O<_-H! N<_  -    d-Experiment settings  =Task graphs: Task number ( N ), shape parameter ( alpha ), average computation cost ( cc ), communication to computation ratio ( CCR ) Best case execution time matrix beta is generated taking into account task heterogeneity and machine heterogeneity b_ij - the best case execution time of task v_i on proc p_j Uncertainty level: degree of uncertainty of actual execution time. UL_ij- the uncertainty level of the execution time of task i on processor j The real execution time is a uniformly distributed random variable The graph has an average uncertainty level UL=C5r"4C@     e.?Bi-objective optimization improves both makespan and robustness@@ @  h/<Relaxing e improves robustness    k0Review  /Want to schedule a DAG on a set of resources Resources may be shared, or variable performance Thus,for each resource, a task has a distribution for its execution time on that resource Want a bounded makespan for the DAG Scheduled performance should be relatively stable with respect to the expected makespan. Schedule is statically generated (not modified dynamically during runtime) The schedule should, as much as possible, be able to withstand variations in the execution times of the individual tasks Optimize the slack subject to constraints on makespan`-$T-$+$ 0 S$ Conclusions  We developed an algorithm for scheduling DAG-structured applications with goals of both minimizing the makespan and maximizing the robustness. Due to conflicting of the two goals, epsilon-constraint method is used to solve the bi-objective optimization problem. Proposed two definitions of robustness Slack is an effective metric to be used to adjust the robustness. The algorithm is flexible in finding the epsilon value in certain user provided range so that the best overall performance is achieved.  L GridSolve Update  GridSolve 0.15 released 05/2006 Support for NATs and firewalls Based on GridRPC Support for batch queues Improved problem descriptions gsIDL Matlab, C, Fortran client interfaces History based execution models Scheduling using perturbation model Win32 native client Lots of work left Client bindings (Mathematica, IDL, ...) Service libraries (ScaLAPACK, ARPACK, ...) Backend resource managers (Condor, VGrADS, ...)L  t,  V        ? m1 FT-MPI Update  FT-MPI: Version 1.01 Fast, scalable fault tolerant MPI implementation System reports failures, recovers processes and messages User must recover program (via checkpoint/restart, ...) Status update In maintenance mode Improvements in stability, performance:;;  w3Open MPI Update  Stable MPI-2 library  version 1.1 A mixture of ideas coming from previous MPI libraries (FT-MPI, LA-MPI, LAM, PACX-MPI) Team: Indiana U, U Tennessee, LANL, HLRS, U Houston, Cisco, Voltaire, Mellanox, Sun, Sandia, Myricom, IBM Multiple simultaneous transports (ethernet, myrinet, ...) Resource managers (ssh, BProc, PBS, XGrid, Slurm, ...) Production quality, thread safe, high performance Tuned collective operations Fault tolerance Involuntary coordinated checkpointing Application unaware that it was checkpointed (by SC06) FT-MPI technologies New FT framework in progress#Xj&7#Xj&  7      (             M!The End   / NOU V!X#Y$]%_'a(c)f*g+i,l.o/x1b 1  h(  hr h S Pi#  i r h S  iy i ^ h 6A8c?``H h 0޽h ? lbrĄw"o{9N(/ 0DArialgbatl?\|<% 0a 0 "DHelveticaatl?\|<% 0a 04 " DTimesicaatl?\|<% 0a 04 0DComic Sans MS\|<% 0a 0 B@DZapf Dingbats\|<% 0a 0PDCalibrigbats\|<% 0a 0"`DGenevagbats\|<% 0a 0pDOttawagbats\|<% 0a 0"DSymbolgbats\|<% 0a 0 c0. @n?" dd@  @@``    .&  &&)/ )x G    _ ]$$ %   b * p ! # b$ٮ5FE3]u7zn u=b$xa"NTyݨIl u=b$cmSұ 4eBP=b$-qqԤM1+=b$,:u^J] 3=b$H8$^3CG@=b$ono * C=b$4!+tCSD P=b$VJ^]/o]u0 _\=$=b$8sb3ef=$=b$Wh~t6 )h=b$gL,)i)Jx _q=$=$=$=b$(%QnO!|=$=b$M&Jok;1=b$LSi=$=b$_S盛HIruh\"=$=b$o' 59%~=$=R$ P߶\\y*4|D=$=b$Q³f#*^[/Q#=b$Pr;vkmh 1=$=b$j*f78+7Y,?=b$b!0nDR5h2 @=b$UF5*$ n&b/`=$=$=$=$=b$ܮ$mf\y*c=$=b$Q3N8Lkؾ>|=# l8c?`!3M@8FG+ Hʚ;y0ʚ;<4dddd= 0g4=d=d% 0hppp@ <4!d!d 0\h=g4PdPd% 0p@ pp <4BdBdf 0\9)___PPT12 %h___PPT2001D<4X@___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography 2DEFAULTFONTSIZE10.DEFAULTWIDTH4000DEFAULTHEIGHT3380___PPT10 ?  O  0LRobust Task Scheduling in Non-deterministic Heterogeneous Computing SystemsM M M qZhiao Shi Asim YarKhan, Jack Dongarra Followed by GridSolve, FT-MPI, Open MPI updates VGrADS Workshop Sept 20060r  h4    L H"General Task Scheduling Problem  Task scheduling: Allocation of tasks of a parallel program to processors to optimize certain goals, e.g. overall execution time (makespan) Application model: task graph (DAG) Node  computational task, has an execution cost Edge  dependency between tasks, data transfer Computing system model Network of processing element (processor memory, unit, communication via message-passing) Optimal task scheduling problem is NP-complete Heuristics: polynomial Some assume every task has same computation cost, (homogenous tasks), some assume arbitrary cost Some ignore communication cost$bZ/$bZ  /$  i I#Robust Static Task Scheduling  Heterogeneous and non-deterministic resources Application DAG with task execution distributions Traditional goal: Minimize the overall makespan based on expected system performance Our Goal: Find static schedules that are more robust to varying task execution time Scheduled performance should be relatively stable with respect to the expected makespan. Approach: Use  slack to absorb task execution time increases caused by uncertainties Employ genetic algorithm for optimization` Z wMZ w  \)6Robustness  definition (I)  fRelative schedule tardiness : Makespan of the schedule obtained with expected task execution time : Real makespan with realization of task execution time Each realization of expected values gives different schedule tardiness Robustness 1 Reflects the amount by which the realized makespan exceeds the expected makespan` Q Q   g Q&8Robustness  definition (II)  Schedule miss rate Robustness 2 A simpler definition, simply counts the number of times the realized makespan exceeds the expected makespan <%m m  R'Calculating Makespan  Given workflow, processors, and a schedule (assign tasks to processors) Adjust communication costs Makespan = longest path from source to sink  `+Defining Slack   JSlack of a task node i is defined as follow: - bottom level (longest path from exit node to node i ) - top level (longest path from entry node to node i) Average slack of a schedule Usefulness of slack in improving robustness Slack at each task of a schedule reflects the  wiggle-room that task has. Large slack means a task node can tolerate large increase of execution time without increasing the makespan L/vI/vI  b,Bi-objective optimization  (Want to optimize both makespan and robustness (as represented by slack) Turn out to be conflicting goals e-constraint Optimize one objective, subject to constraints imposed on the other objectives maximize average slack of the schedule ( ) subject to: Where HEFT is a well known, efficient algorithm for serializing a DAG and assigning a schedule Use genetic algorithm to do the optimization Fitness function is average slack For solution violating constraint, fitness is penalized by the degree of violation of the constraint H! O<_-H! N<_  -    d-Experiment settings  =Task graphs: Task number ( N ), shape parameter ( alpha ), average computation cost ( cc ), communication to computation ratio ( CCR ) Best case execution time matrix beta is generated taking into account task heterogeneity and machine heterogeneity b_ij - the best case execution time of task v_i on proc p_j Uncertainty level: degree of uncertainty of actual execution time. UL_ij- the uncertainty level of the execution time of task i on processor j The real execution time is a uniformly distributed random variable The graph has an average uncertainty level UL=C5r"4C@     e.?Bi-objective optimization improves both makespan and robustness@@ @  h/<Relaxing e improves robustness    k0Review  /Want to schedule a DAG on a set of resources Resources may be shared, or variable performance Thus,for each resource, a task has a distribution for its execution time on that resource Want a bounded makespan for the DAG Scheduled performance should be relatively stable with respect to the expected makespan. Schedule is statically generated (not modified dynamically during runtime) The schedule should, as much as possible, be able to withstand variations in the execution times of the individual tasks Optimize the slack subject to constraints on makespan`-$T-$+$ 0 S$ Conclusions  We developed an algorithm for scheduling DAG-structured applications with goals of both minimizing the makespan and maximizing the robustness. Due to conflicting of the two goals, epsilon-constraint method is used to solve the bi-objective optimization problem. Proposed two definitions of robustness Slack is an effective metric to be used to adjust the robustness. The algorithm is flexible in finding the epsilon value in certain user provided range so that the best overall performance is achieved.  L GridSolve Update  GridSolve 0.15 released 05/2006 Support for NATs and firewalls Based on GridRPC Support for batch queues Improved problem descriptions gsIDL Matlab, C, Fortran client interfaces History based execution models Scheduling using perturbation model Win32 native client Lots of work left Client bindings (Mathematica, IDL, ...) Service libraries (ScaLAPACK, ARPACK, ...) Backend resource managers (Condor, VGrADS, ...)L  t,  V        ? m1 FT-MPI Update  FT-MPI: Version 1.01 Fast, scalable fault tolerant MPI implementation System reports failures, recovers processes and messages User must recover program (via checkpoint/restart, ...) Status update In maintenance mode Improvements in stability, performance:;;  w3Open MPI Update  Stable MPI-2 library  version 1.1 A mixture of ideas coming from previous MPI libraries (FT-MPI, LA-MPI, LAM, PACX-MPI) Team: Indiana U, U Tennessee, LANL, HLRS, U Houston, Cisco, Voltaire, Mellanox, Sun, Sandia, Myricom, IBM Multiple simultaneous transports (ethernet, myrinet, ...) Resource managers (ssh, BProc, PBS, XGrid, Slurm, ...) Production quality, thread safe, high performance Tuned collective operations Fault tolerance Involuntary coordinated checkpointing Application unaware that it was checkpointed (by SC06) FT-MPI technologies New FT framework in progress#Xj&7#Xj&  7      (             M!The End   / NOU V!X#Y$]%_'a(c)f*g+i,l.o/x1r#"#{Root EntrydO)2 GPicturescCurrent UserMSummaryInformation(XR J    !"#$%&'()*+,-./0123456789:;<=>?@ABCDEHKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz|}~ Design Template Slide Titles'_#|Charles KoelbelCharles Koelbel